Water pump motor temperature estimation method based on data and model fusion

Abstract

The invention discloses a water pump motor temperature estimation method based on data and model fusion. The method comprises the following steps: obtaining multi-source operation data of a water pump motor under different operation conditions; constructing a multi-node thermal network model of the motor for the water pump, and calculating and outputting multi-node temperature estimation values under different operation conditions based on the multi-source operation data; constructing an LSTM neural network model, and introducing physical prior constraints to obtain a composite loss function fused with physical prior; training an LSTM neural network model, and optimizing and adjusting the LSTM neural network model by using a composite loss function to obtain a temperature prediction model; when the temperature of the motor for the water pump is estimated, motor operation data under the corresponding operation working condition are input into the temperature prediction model, and the temperature of each hot node of the motor for the water pump is output. According to the method, the prior structure information provided by the thermal network model can be utilized, the capability of self-adaptive modeling for different working conditions is achieved, and efficient and accurate estimation of the internal temperature of the motor for the water pump is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of motor temperature estimation and thermal management, and more particularly to a method for estimating the temperature of a water pump motor based on data and model fusion. Background Art

[0002] As a key device in fluid conveying systems, the operating efficiency and stability of water pumps directly affect energy consumption and system reliability in various industrial and civilian applications. Currently, water pumps are widely driven by high-efficiency motors. These motors typically operate under high loads and long operating cycles. The working environment is often accompanied by problems such as enclosed spaces, high humidity, and poor heat dissipation, which makes the motor prone to high temperature rise during operation. Once the motor temperature rise exceeds the reasonable range, it will not only accelerate the aging of the winding insulation and the degradation of magnetic properties, but may also trigger thermal overload protection and even burn the equipment, seriously affecting the safe operation and service life of the entire water pump system.

[0003] During actual operation, the internal structure of water pump motors is typically compact, making it difficult to measure the temperature of some key components in real time using traditional contact temperature sensors. This is especially true in the high-speed rotating rotor, where traditional temperature measurement methods make it nearly impossible to deploy sensors. Existing wireless temperature measurement solutions are costly, lack stability, and can damage the motor's internal structure during installation, impacting normal operation.

[0004] Currently, mainstream methods for motor temperature estimation can be divided into the following categories: The first category involves estimation methods based on thermal physics modeling, such as thermal network models or finite element thermal simulation models. These methods dynamically estimate the internal temperature by modeling the motor's heat source distribution, structural parameters, and material thermal properties, combined with the solution of thermal equilibrium differential equations. These methods offer good physical interpretability and versatility, but they are highly dependent on model parameters, complex in the modeling process, and difficult to adapt to variable operating conditions and operating disturbances, limiting their practical applicability. The second category involves data-driven intelligent approaches, typically employing machine learning algorithms such as neural networks, support vector machines, and random forests. These methods utilize large amounts of operating data for offline training to establish a mapping between motor operating variables and internal temperature. These methods can achieve high prediction accuracy when sufficient data is available and operating conditions are relatively fixed. However, due to the lack of physical constraints, these models have limited generalization capabilities to unseen operating conditions. Furthermore, they exhibit "black box" properties, making it difficult to interpret the physical meaning of the estimation results. Furthermore, they place high demands on the size and quality of the training samples.

[0005] In summary, it is difficult for existing methods to balance the accuracy, robustness, and practicality of the temperature estimation model in actual water pump application scenarios where motor sensor layout is limited and the operating environment is complex and changeable.

[0006] Therefore, providing a temperature estimation method that integrates physical modeling and data learning capabilities is an urgent problem that needs to be solved by those skilled in the art. Summary of the Invention

[0007] In view of this, the present invention provides a temperature estimation method for water pump motors based on data and model fusion. It combines the motor thermal network model with the neural network, constructs low-recognition samples through the thermal network and guides the neural network training. At the same time, a small amount of measured temperature data is used to fine-tune the model, thereby achieving high-precision estimation of the internal temperature of the permanent magnet synchronous motor for water pumps.

[0008] In order to achieve the above object, the present invention adopts the following technical solutions:

[0009] In a first aspect, the present invention provides a method for estimating the temperature of a water pump motor based on data and model fusion, comprising the following steps:

[0010] S1. Obtain multi-source operating data of water pump motors under different operating conditions and construct a time series dataset;

[0011] S2. Construct a multi-node thermal network model for the water pump motor, and calculate and output multi-node temperature estimation values ​​under different operating conditions based on the multi-source operating data;

[0012] S3. Build an LSTM neural network model and introduce the multi-node temperature estimates under the corresponding operating conditions as physical prior constraints in the loss function to obtain a composite loss function that incorporates physical priors.

[0013] S4. Training the LSTM neural network model based on the multi-node temperature estimation values ​​under the different operating conditions, learning the temporal mapping relationship between the time series data set and the temperature changes of each hot node, and optimizing and adjusting the LSTM neural network model using the composite loss function to obtain a temperature prediction model;

[0014] S5. When estimating the temperature of the water pump motor, the motor operation data under the corresponding operating conditions is input into the temperature prediction model, and the temperature of each thermal node of the water pump motor is output.

[0015] Furthermore, the multi-source operating data includes stator current, voltage and ambient temperature.

[0016] Furthermore, step S2 specifically includes:

[0017] S21. Construct a multi-node thermal network structure for a water pump motor; wherein each node includes: a stator winding, a stator core, a rotor, and a housing; and any two nodes are connected via a thermal resistance;

[0018] S22. Calculate the heating power of each node based on the motor operating data under different operating conditions; obtain the corresponding stator copper loss, core loss, and permanent magnet eddy current loss as heat source input;

[0019] S23, establishing a thermal network dynamic equation group based on the heat capacity of each node according to the multi-node thermal network structure and heat source input;

[0020] S24. Solve the thermal network dynamic equations using the Euler method or the fourth-order Runge-Kutta method to obtain the temperature estimation value of each node under different operating conditions.

[0021] Furthermore, the thermal network dynamic equations are expressed in differential form as follows:

[0022]

[0023] Where i is the number of nodes, C i is the heat capacity of the ith node, R i,j is the thermal resistance between node i and node j, ￥(i) is the set of adjacent nodes directly connected to node i; P i,z is the heat source input of the i-th node under working condition z, T j,z is the estimated temperature of the jth node under working condition z, T i,z is the estimated temperature of the i-th node under working condition z.

[0024] Furthermore, the LSTM neural network model in step S3 includes: an input layer, an LSTM memory layer, and a fully connected output layer;

[0025] The input layer is used to receive motor operation data under corresponding working conditions, and is organized as a sequence input in a sliding window manner in the time dimension;

[0026] The LSTM memory layer includes multiple LSTM units, each of which is used to extract the time correlation and dynamic change trend in the input sequence, thereby obtaining the nonlinear characteristics of the temperature change of the water pump motor with the operating conditions;

[0027] The fully connected output layer maps the temporal feature representation of the LSTM memory layer into temperature estimates of multiple nodes of the motor.

[0028] Furthermore, the composite loss function integrating physical priors in step S3 is expressed as follows:

[0029] L total =L data +λ·L phys

[0030]

[0031] Among them, L data is the data fitting error term, L phys is the physical consistency error term, λ is the weight factor, is the temperature prediction value of the kth sample, is the temperature estimate of the kth sample, Y k is the measured temperature value of the kth sample; N is the total number of samples.

[0032] Furthermore, step S4 includes:

[0033] S41, based on the motor operating data under different operating conditions and the corresponding multi-node temperature estimation values ​​in the multi-node thermal network model, training the LSTM neural network model; fully learning the time series mapping relationship between the input stator current, voltage, ambient temperature and the temperature change of each thermal node;

[0034] S42. Optimize network parameters by back propagation using the composite loss function, and use the converged model as a preliminary temperature prediction model.

[0035] Furthermore, step S4 further includes:

[0036] S43. Using the measured temperature data of the water pump motor under different working conditions as a fine-tuning set, adjusting the output layer parameters of the preliminary temperature prediction model based on the composite loss function to obtain a final temperature prediction model.

[0037] It can be seen from the above technical solution that, compared with the prior art, the present invention discloses a method for estimating the temperature of a water pump motor based on data and model fusion, which has the following beneficial effects:

[0038] First, by constructing a multi-node thermal network model for water pump motors, this model can reflect the thermal coupling characteristics of key motor components, such as the stator winding, core, and rotor. This provides a physically consistent temperature estimation basis and low-precision data in scenarios where internal sensors are lacking. The output of this thermal network model is then introduced as a physical prior into the neural network structure, constructing a neural network model with physical constraints. This ensures that the network maintains its output consistent with thermal physical laws while learning, thereby improving the credibility and physical interpretability of the neural network estimation results. Finally, the parameters of the neural network output layer are fine-tuned using actual water pump motor operating data to obtain a water pump motor temperature estimation model. This integrated approach effectively improves the estimation accuracy and response speed of the temperature of key internal motor components, meeting the practical needs of online thermal status monitoring and dynamic control of water pump motors.

[0039] This method utilizes a thermal network model to construct multi-node thermal behavior characteristics for water pump motors. By incorporating this model output as a priori constraints into a neural network model, it achieves highly accurate temperature estimation. This method improves the real-time and accuracy of water pump motor thermal status monitoring, effectively overcoming the accuracy bottleneck of traditional temperature estimation methods and meeting the practical need for dynamic estimation of the internal temperature of water pump motors under complex operating conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0041] Figure 1 A flow chart of a method for estimating the temperature of a water pump motor based on data and model fusion provided in an embodiment of the present invention.

[0042] Figure 2 Schematic diagram of a four-node thermal network model for a water pump motor provided in an embodiment of the present invention.

[0043] Figure 3 This is a comparison chart of the predicted winding temperature output by the temperature prediction model provided by an embodiment of the present invention and the actual measured winding temperature.

[0044] Figure 4 This is a comparison chart of the rotor predicted temperature output by the temperature prediction model provided by an embodiment of the present invention and the rotor measured temperature. DETAILED DESCRIPTION

[0045] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0046] The embodiment of the present invention discloses a method for estimating the temperature of a water pump motor based on data and model fusion, referring to Figure 1 As shown, the following steps are included:

[0047] S1. Obtain multi-source operating data of water pump motors under different operating conditions and construct a time series dataset;

[0048] S2. Construct a multi-node thermal network model for the water pump motor, and calculate and output multi-node temperature estimation values ​​under different operating conditions based on the multi-source operating data;

[0049] S3. Build an LSTM neural network model and introduce the multi-node temperature estimates under the corresponding operating conditions as physical prior constraints in the loss function to obtain a composite loss function that incorporates physical priors.

[0050] S4. Training the LSTM neural network model based on the multi-node temperature estimation values ​​under the different operating conditions, learning the temporal mapping relationship between the time series data set and the temperature changes of each hot node, and optimizing and adjusting the LSTM neural network model using the composite loss function to obtain a temperature prediction model;

[0051] S5. When estimating the temperature of the water pump motor, the motor operation data under the corresponding operating conditions is input into the temperature prediction model, and the temperature of each thermal node of the water pump motor is output.

[0052] When estimating the temperature of the water pump motor, the present invention can not only utilize the prior structural information provided by the thermal network model, but also has the ability to adaptively model different working conditions, thereby achieving efficient and accurate estimation of the internal temperature of the water pump motor and improving the safety and intelligence level of the system operation.

[0053] This embodiment is described by taking the estimation of the internal temperature of a permanent magnet synchronous motor body as an example.

[0054] Step 1: Obtain the operating data of the water pump motor: Collect multi-source operating data of the water pump permanent magnet synchronous motor under different operating conditions, including stator current, voltage, ambient temperature, etc., and construct a time series operating data set in a unified format;

[0055] Step 2: Build a multi-node thermal network model for the water pump motor: Based on the principle of heat conduction, a motor thermal network model is constructed that includes key nodes such as the stator winding, stator core, and rotor. The model calculates and outputs multi-node temperature estimates under different operating conditions to form a low-precision temperature estimation dataset.

[0056] Step 3: Build a neural network model for estimating motor temperature. Introduce the temperature prediction value from the motor thermal network model as a physical prior constraint in the neural network's loss function. Construct a composite loss function that includes the prediction error of the actual temperature measurement label and the consistency error of the neural network output with the motor thermal network prediction, thereby improving the model's physical consistency with thermal behavior.

[0057] Step 4: Train the neural network model built in S3: Use the low-precision data in S2 and the operating condition data combination sequence to train the neural network model. The model optimizes the parameters through back propagation so that it can fully learn the time series mapping relationship between the input operating condition sequence and the temperature changes of each thermal node, and obtain a temperature prediction model with basic estimation capabilities;

[0058] Step 5: Fine-tune the model to improve the accuracy of internal temperature estimation of the water pump motor: Use a small amount of measured temperature data in S1 to fine-tune the neural network model, adjusting only the parameters of the output layer to improve the model's temperature estimation accuracy in the water pump system.

[0059] This embodiment of the water pump permanent magnet motor temperature estimation method uses a thermal network model to construct multi-node thermal behavior characteristics for the water pump motor. This model output is then incorporated into the neural network model as a priori constraints, achieving highly accurate temperature estimation. This method improves the real-time and accuracy of water pump motor thermal status monitoring, effectively overcoming the accuracy bottleneck of traditional temperature estimation methods and meeting the practical needs for dynamic estimation of the internal temperature of water pump motors under complex operating conditions.

[0060] The following describes the various steps of this embodiment in detail.

[0061] Step 1: Obtain the operating data of the water pump motor.

[0062] Various sensors, such as current sensors, voltage sensors, and temperature sensors, are used to collect multi-source operating data of permanent magnet synchronous motors for water pumps under different operating conditions. This data includes stator current, voltage, torque, speed, and ambient temperature. For example, the current of the DC and Q axes of the water pump motor in the rotating coordinate system, the control voltage of the DC and Q axes of the motor, the rotor angular velocity, the motor torque, and the ambient temperature. This data constitutes a time series dataset, and some data includes temperature labels for different components within the motor at the corresponding time.

[0063] Step 2: Build a multi-node thermal network model for the water pump motor; including:

[0064] Step 2.1: Construct the multi-node thermal network structure of the motor.

[0065] This embodiment is based on the principle of heat conduction to establish an N-point thermal network model for the water pump motor. Figure 2 As shown, each node may include the stator winding (node ​​1), the stator core (node ​​2), the rotor (node ​​3), the housing environment, etc. This embodiment constructs a four-node thermal network model, where any two nodes are connected by thermal resistance according to actual conditions, and each node corresponds to a certain thermal capacity; Figure 2 Medium R A,2 represents the thermal resistance between the environment and node 2; R 2,1 represents the thermal resistance between node 2 and node 1; R2,3 represents the thermal resistance between node 2 and node 3; R A,3 represents the thermal resistance between the environment and node 3; each node is connected in parallel with a heat capacitor; C1 represents the heat capacitance of node 1; C2 represents the heat capacitance of node 2; and C3 represents the heat capacitance of node 3. The heat capacitance and heat source input of each node are defined. In this embodiment, the node temperature vector T is set as:

[0066] T=[T1,T2,T3,T A ]

[0067] Where, T1, T2 and T3 represent the temperature of the stator winding, stator core and rotor of the water pump motor respectively, T A Indicates the external ambient temperature.

[0068] Step 2.2, define the heat source input.

[0069] The heating power input of each node under the corresponding working conditions is calculated based on the motor operating data (such as current, voltage, speed, etc.).

[0070] The heat source items in this embodiment include:

[0071] Stator copper loss P1:

[0072] P1=1.5i 2 R s

[0073] Where i is the total current of the motor, R s is the motor coil resistance.

[0074] Core loss P2:

[0075] P2=a hys fB 2 +a ext (fB) 3 / 2

[0076] Where a hys 、a ext are the hysteresis loss coefficient and additional loss coefficient respectively, f is the motor frequency, and B is the motor magnetic flux density.

[0077] The eddy current loss P3 of the permanent magnet can be approximated by referring to the technical manual.

[0078] After synthesis, the heat source input vector P is obtained z :

[0079] P z =[P 1,z ,P 2,z ,P 3,z ]

[0080] Where, P1,z 、P 2,z 、P 3,z are the heat sources of motor nodes 1, 2, and 3 under different working conditions z.

[0081] Step 2.3, establish the thermal network dynamic equations.

[0082] Based on the relationship between heat conduction and node heat balance, the following differential form thermal network model is established:

[0083]

[0084] Where i is the number of nodes, C i is the heat capacity of the ith node, R i,j is the thermal resistance between node i and node j, ￥(i) is the set of adjacent nodes directly connected to node i; P i,z is the heat source input of the i-th node under working condition z, T j,z is the estimated temperature of the jth node under working condition z, T i,z is the estimated temperature of the i-th node under working condition z.

[0085] Step 2.4, simulation calculation and low-precision temperature data set generation Under different operating conditions z, the corresponding heat source vector and initial temperature boundary conditions are used to numerically solve the above differential equations using the Euler method or the fourth-order Runge-Kutta method. The solution function is f TN , obtain the steady-state or dynamic temperature solution of each node:

[0086] T z =[T 1,z ,T 2,z ,T 3,z ,T A,z ]=f TN (Z)

[0087] Finally, a low-precision temperature dataset estimated by the thermal network is constructed:

[0088] X LP ={T1,T2,T3,T A}

[0089] This dataset will be used for pre-training of subsequent data-driven models or as a source of physical constraints for hybrid modeling.

[0090] Step 3: Build a neural network model for estimating motor temperature and construct a composite loss function that incorporates physical priors. This includes:

[0091] Step 3.1, input data preparation and neural network construction.

[0092] The neural network model selected in this embodiment is a long short-term memory artificial neural network (LSTM). In this embodiment, the constructed LSTM neural network model adopts a typical time series prediction structure. The whole is composed of an input layer, an LSTM memory layer and a fully connected output layer, and a physical constraint module is integrated to enhance the rationality and interpretability of the prediction. The input layer is used to receive the processed motor operation data, which is organized as a sequence input in a sliding window manner in the time dimension.

[0093] The collected motor operation data is constructed as a time series sample. Assuming the input feature dimension is S and the time step is T, the input of the kth sample is recorded as:

[0094]

[0095] Where, Represents the input vector of the kth sample at the Tth time.

[0096] The LSTM memory layer consists of multiple LSTM units. Each unit includes a forget gate, an input gate, an output gate, and a unit state. These units are used to extract the temporal correlation and dynamic change trends in the input sequence, thereby characterizing the nonlinear characteristics of the water pump motor temperature as it changes with operating conditions.

[0097] The fully connected output layer maps the time series feature representation of the LSTM layer into temperature estimates of multiple key parts of the motor. The LSTM network is used to learn the nonlinear mapping relationship between the input sequence and the target temperature output. Its output is:

[0098]

[0099] Where θ is the network parameter to be trained, is the temperature prediction value of the kth sample.

[0100] In step 3.2, the motor thermal network model output is introduced as a physical prior.

[0101] Using the motor thermal network model constructed in step 2, based on the same input, calculate the low-precision temperature estimate corresponding to the kth sample As a physical prior.

[0102]

[0103] Where, F TN Solve function f for motor temperature TN A collection of time series.

[0104] Step 3.3: Construct a composite loss function that integrates physical priors.

[0105] The goal of the LSTM neural network model is not only to minimize the difference with the true label, but also to maintain physical consistency with the model's prediction results. To this end, this embodiment constructs a composite loss function as follows:

[0106] L total =L data +λ·L phys

[0107] Among them, the data fitting error term (mean square error) is:

[0108]

[0109] The physical consistency error term is:

[0110]

[0111] Among them, L data is the data fitting error term, L phys is the physical consistency error term, is the temperature prediction value of the kth sample, is the temperature estimate of the kth sample, Y k is the measured temperature value of the kth sample; N is the total number of samples; and λ is a weighting factor that adjusts the relative importance of the two loss terms. In this example, λ = 0.2. This loss function ensures the model's ability to fit real temperature measurement data while preserving the physical consistency of the motor's thermal behavior.

[0112] In step 3.4, the neural network is trained based on the composite loss function, using stochastic gradient descent to iteratively update the parameter θ to minimize it. During training, the model uses physical guidance to suppress overfitting and improve the reliability of the estimate in the absence of high-precision samples.

[0113] In step 4, by using the temperature data generated by the thermal network model under different working conditions, the LSTM neural network model is trained to learn the time series mapping relationship between the input variables and the motor temperature, and a prediction model with preliminary temperature estimation capabilities is constructed.

[0114] This embodiment specifically includes the following steps:

[0115] Step 4.1: Organize the data generated by the thermal network model into a fixed time series as the model input, and use the temperature output by the thermal network model as the training target;

[0116] Step 4.2: Build a neural network structure containing several layers of LSTM and set training parameters such as learning rate, batch size, and number of training rounds.

[0117] Step 4.3: Perform model training based on the composite loss function in S3 and optimize network parameters through backpropagation.

[0118] In step 4.4, the training effect is evaluated on the validation set, and the converged model is saved as the temperature estimation model.

[0119] Step 5: Fine-tune the model to improve the accuracy of internal temperature estimation of the water pump motor.

[0120] To improve the temperature estimation accuracy of the model under actual working conditions, this embodiment uses a small amount of labeled measured temperature data to fine-tune the trained LSTM model. Specifically, the following steps are performed:

[0121] Step 5.1: Select a small-scale measured data set containing high precision as the fine-tuning set;

[0122] Step 5.2: Freeze all parameters except the output layer in the constructed LSTM model and only perform gradient updates on the output layer parameters.

[0123] Step 5.3: Use the measured temperature as the supervision target and optimize using the composite loss function;

[0124] Through this step, the model enhances its ability to fit actual samples while maintaining the original physical structure constraints, significantly improving the estimation accuracy and practical application effect.

[0125] To estimate the temperature of a water pump motor, the first step is to collect the actual operating data of the water pump's permanent magnet synchronous motor. This data is collected in real time through embedded sensors or control systems (such as PLCs, inverters, and controllers) including stator current, voltage, ambient temperature, load status, speed, and other relevant sensor data, and formatted in a unified format. This motor operating data is then fed into the final temperature prediction model. The model, based on the input operating condition sequence and combining learned temporal mapping relationships with prior physical knowledge, outputs the temperature of each thermal node in the water pump motor.

[0126] The comparison chart of the winding predicted temperature output by the temperature prediction model of this embodiment and the winding measured temperature is shown in FIG. Figure 3 As shown, the comparison chart of the rotor predicted temperature and the rotor measured temperature is shown in Figure 4As shown, it shows that the predicted temperature and the measured temperature are basically consistent during the monitoring period. The temperature estimation method of the present invention can improve the real-time and accuracy of the thermal state estimation of the water pump motor. It integrates the dual advantages of the physical model and the data model, and can accurately predict the temperature distribution of key components inside the motor when the sensor is inaccessible and the working conditions are complex and changeable. It has important theoretical research value and engineering application potential. It can not only utilize the prior structure information provided by the thermal network model, but also has the ability to adaptively model different working conditions to achieve efficient and accurate estimation of the internal temperature of the water pump motor, thereby improving the safety and intelligence level of the system operation.

[0127] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0128] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for estimating the temperature of a water pump motor based on data and model fusion, characterized in that: The following steps are involved: S1. Obtain multi-source operating data of water pump motors under different operating conditions and construct a time series dataset; S2. Construct a multi-node thermal network model for the water pump motor, and calculate and output multi-node temperature estimation values ​​under different operating conditions based on the multi-source operating data; S3. Build an LSTM neural network model and introduce the multi-node temperature estimates under the corresponding operating conditions as physical prior constraints in the loss function to obtain a composite loss function that incorporates physical priors. S4. Training the LSTM neural network model based on the multi-node temperature estimation values ​​under the different operating conditions, learning the temporal mapping relationship between the time series data set and the temperature changes of each hot node, and optimizing and adjusting the LSTM neural network model using the composite loss function to obtain a temperature prediction model; S5. When estimating the temperature of the water pump motor, the motor operation data under the corresponding operating conditions is input into the temperature prediction model, and the temperature of each thermal node of the water pump motor is output.

2. A method for estimating temperature of a water pump motor based on data and model fusion according to claim 1, characterized in that: The multi-source operating data includes stator current, voltage, torque, speed and ambient temperature.

3. The method for estimating the temperature of a water pump motor based on data and model fusion according to claim 1, wherein: Step S2 specifically includes: S21. Construct a multi-node thermal network structure for a water pump motor; wherein each node includes: a stator winding, a stator core, a rotor, and a housing; and any two nodes are connected via a thermal resistance; S22. Calculate the heating power of each node based on the motor operating data under different operating conditions; obtain the corresponding stator copper loss, core loss, and permanent magnet eddy current loss as heat source input; S23, establishing a thermal network dynamic equation group based on the heat capacity of each node according to the multi-node thermal network structure and heat source input; S24. Solve the thermal network dynamic equations using the Euler method or the fourth-order Runge-Kutta method to obtain the temperature estimation value of each node under different operating conditions.

4. A method for estimating the temperature of a water pump motor based on data and model fusion according to claim 3, characterized in that: The thermal network dynamic equations are expressed in differential form as follows: Where i is the number of nodes, C i is the heat capacity of the ith node, R i,j is the thermal resistance between node i and node j, ￥(i) is the set of adjacent nodes directly connected to node i; P i,z is the heat source input of the i-th node under working condition z, T j,z is the estimated temperature of the jth node under working condition z, T i,z is the estimated temperature of the i-th node under working condition z.

5. The method for estimating the temperature of a water pump motor based on data and model fusion according to claim 1, wherein: The LSTM neural network model in step S3 includes: an input layer, an LSTM memory layer, and a fully connected output layer; The input layer is used to receive motor operation data under corresponding working conditions, and is organized as a sequence input in a sliding window manner in the time dimension; The LSTM memory layer includes multiple LSTM units, each of which is used to extract the time correlation and dynamic change trend in the input sequence, thereby obtaining the nonlinear characteristics of the temperature change of the water pump motor with the operating conditions; The fully connected output layer maps the temporal feature representation of the LSTM memory layer into temperature estimates of multiple nodes of the motor.

6. The method for estimating the temperature of a water pump motor based on data and model fusion according to claim 1, wherein: The composite loss function integrating physical priors in step S3 is expressed as follows: L total =L data +λ·L phys Among them, L data is the data fitting error term, L phys is the physical consistency error term, λ is the weight factor, is the temperature prediction value of the kth sample, is the temperature estimate of the kth sample, Y k is the measured temperature value of the kth sample; N is the total number of samples.

7. The method for estimating the temperature of a water pump motor based on data and model fusion according to claim 3, characterized in that: Step S4 includes: S41, based on the motor operating data under different operating conditions and the corresponding multi-node temperature estimation values ​​in the multi-node thermal network model, training the LSTM neural network model; fully learning the time series mapping relationship between the input stator current, voltage, ambient temperature and the temperature change of each thermal node; S42. Optimize network parameters by back propagation using the composite loss function, and use the converged model as a preliminary temperature prediction model.

8. The method for estimating the temperature of a water pump motor based on data and model fusion according to claim 7, characterized in that: Step S4 further includes: S43. Using the measured temperature data of the water pump motor under different working conditions as a fine-tuning set, adjusting the output layer parameters of the preliminary temperature prediction model based on the composite loss function to obtain a final temperature prediction model.

Images