A control method for multi-motor cooperation of intelligent multiphase booster pump device

Abstract

The application relates to the technical field of intelligent pump control, and discloses a control method for multi-motor cooperation of an intelligent multiphase booster pump device, which comprises the following steps: performing format specification and interference elimination processing on historical operation data and real-time working condition data of the multiphase booster pump device to obtain a standardized operation data set; mapping the standardized operation data set to a state space to generate a motor state mapping vector and a working condition mapping vector; calibrating distribution differences of the motor state mapping vector to obtain a unified representation of multi-motor cooperation; taking the unified representation and real-time data of a target pump group as inputs and taking motor control instructions as outputs to generate motor control instructions for current operation of the target pump group; and calculating a pump group level control threshold value, combining the motor control instructions for current operation of the target pump group, and judging whether to trigger control adjustment. The control method can solve the problem that existing pump device control methods are difficult to dynamically calibrate the cooperative relationship among multiple motors. The control method can improve the overall operation coordination of the pump group and the safety of the system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent pump control, in particular to a control method for multi-motor cooperation of an intelligent multiphase booster pump device. BACKGROUND

[0002] In the industrial fields of petroleum, chemical industry, water treatment, etc., the multiphase booster pump device, as the core equipment of the fluid conveying system, directly affects the safety and economy of the production process in terms of operation efficiency and stability. The multiphase booster pump device is composed of multiple motor-driven pump groups, which realize medium boosting, flow regulation and pressure control under complex working conditions through cooperative work. However, due to the inherent differences in manufacturing process, running loss, installation position, etc. of different motors, combined with the dynamic changes of working conditions (such as medium composition, temperature, pipeline pressure), it is easy to cause uneven load distribution and inconsistent response delay among motors, and further cause risks such as excessive local energy consumption, accelerated equipment wear and even system shutdown.

[0003] The traditional multi-motor control method mostly adopts a single motor independent adjustment strategy, that is, through collecting the speed, current and other parameters of a single motor for closed-loop control, lacking overall consideration of the cooperative relationship among multiple motors. This way can basically meet the operation requirements when the working condition is stable, but in complex dynamic scenarios (such as sudden change of medium flow, fluctuation of pipeline pressure), it is easy to cause "load stealing" or "underload" phenomenon among motors. For example, when the load of a motor suddenly rises due to the increase of local resistance in the pipeline, if the output of other motors remains unchanged, it will lead to imbalance of the overall pressure of the pump group, not only reducing the system efficiency, but also possibly causing safety hazards such as pipeline leakage.

[0004] With the development of industrial intelligence, some control methods introduce simple cooperative logic, such as distributing motor load based on a preset proportion, but such methods rely on manual experience to set parameters and are difficult to adapt to changing working conditions. At the same time, the format of historical operation data and real-time working condition data is not unified, and there are many interference noises, etc., which leads to low data utilization rate and cannot provide accurate basis for cooperative control. In addition, there is a lack of deep fusion modeling of motor state and working condition characteristics in the prior art, which makes it difficult to realize unified representation of multi-motor state, and the adaptability and robustness of the cooperative control strategy are insufficient. Therefore, how to construct a control method that can effectively process multi-source data, accurately map the state of the motor, and dynamically calibrate the cooperative relationship of multiple motors has become a key challenge to improve the operation performance of intelligent multiphase booster pump devices. SUMMARY

[0005] The purpose of the present application is to provide a control method for multi-motor cooperation of an intelligent multiphase booster pump device to solve the problems raised in the background.

[0006] To achieve the above object, the application provides a control method for multi-motor cooperation of an intelligent multi-phase booster pump device, which comprises the following steps:

[0007] Collecting historical operation data and real-time working condition data of the multi-phase booster pump device, and performing format specification and interference elimination processing on the two types of data to obtain a standardized operation data set;

[0008] Constructing a multi-motor state mapping module, mapping motor state features and working condition features in the standardized operation data set to a preset state space to generate a motor state mapping vector and a working condition mapping vector;

[0009] Based on the motor state mapping vector and the working condition mapping vector, a multi-motor cooperation calibration model is constructed, and the distribution difference of different motor state mapping vectors is calibrated to obtain a multi-motor cooperation unified representation;

[0010] Training a cooperative control execution model, taking the multi-motor cooperation unified representation and target pump group real-time data as input data, and taking motor control instructions as output data to generate motor control instructions for the current operation of the target pump group;

[0011] According to the instruction distribution of the historical operation data of the target pump group, a pump group level control threshold is calculated, and the motor control instructions for the current operation of the target pump group and the pump group level control threshold are combined to determine whether to trigger control adjustment, and the parameters of the multi-motor cooperation calibration model are updated regularly.

[0012] Preferably, the specific method for collecting historical operation data and real-time working condition data of the multi-phase booster pump device, and performing format specification and interference elimination processing on the two types of data to obtain a standardized operation data set is as follows:

[0013] From the monitoring systems of three types of pump groups, i.e. fluid delivery, pressure regulation and flow distribution, the time mark, motor speed and driving voltage fields in the historical operation data and the medium temperature, pipeline pressure and valve opening degree fields in the real-time working condition data are extracted;

[0014] The extracted fields are subjected to unit unification and missing value completion processing, and the sliding window method is used to filter abnormal fluctuation points in the time mark, and the processed historical operation data and real-time working condition data are respectively stored as structured forms to obtain a standardized operation data set.

[0015] Preferably, the specific method for constructing a multi-motor state mapping module and mapping motor state features and working condition features in the standardized operation data set to a preset state space to generate a motor state mapping vector is as follows:

[0016] A preset state model is selected as a basic mapping model, a state mapping layer is added to the input layer of the basic mapping model, and the structured fields in the standardized operation data set are converted into feature vector form.

[0017] The state extractor is arranged in the middle layer of the basic mapping model, the output of the middle layer is mapped to a state space with a fixed dimension through a linear transformation layer, and the motor state features and the working condition features are respectively encoded to obtain the motor state mapping vector and the working condition mapping vector corresponding to the pump set.

[0018] Preferably, the multi-motor collaborative calibration model is constructed based on the motor state mapping vector and the working condition mapping vector, and the specific method for obtaining the multi-motor collaborative unified representation by calibrating the distribution difference of different motor mapping vectors is as follows:

[0019] An identifier is assigned to each motor, the motor state mapping vector and the working condition mapping vector are taken as input, and a collaborative network including an identifier discriminator and a state calibrator is constructed.

[0020] The identifier discriminator is used to determine the identifier of the motor to which the input motor state mapping vector or working condition mapping vector belongs, and the state calibrator is used to adjust the distribution of the motor state mapping vector or working condition mapping vector, so that the motor state mapping vectors or working condition mapping vectors of different motor identifiers are distributed in the state space.

[0021] By minimizing the classification error of the identifier discriminator and maximizing the calibration error of the state calibrator, the parameters of the multi-motor collaborative calibration model are optimized to obtain the multi-motor collaborative unified representation.

[0022] Preferably, the specific method for training the collaborative control execution model by taking the multi-motor collaborative unified representation and the real-time data of the target pump set as input data and taking the motor control instruction as output data to generate the motor control instruction for the current operation of the target pump set is as follows:

[0023] The historical operation data and real-time working condition data of the target pump set are collected, the multi-motor collaborative unified representation thereof is extracted, the historical operation data is labeled as a control adjustment label, the real-time working condition data is labeled as a normal operation label, and a training sample set is constructed.

[0024] The time sequence features of the multi-motor collaborative unified representation and the real-time data of the target pump set are spliced as input, and a binary classification loss function is used to train the collaborative control execution model, so as to accurately predict the control adjustment label as a training target, and the model training is completed when the loss function converges.

[0025] For the currently running target pump set, the multi-motor collaborative unified representation and the real-time time sequence features thereof are extracted as input data, and the motor control instruction for the current operation of the target pump set is output.

[0026] Preferably, the instruction distribution according to the historical operation data of the target pump set is used to calculate the pump set level control threshold, and the specific method for determining whether to trigger control adjustment is as follows:

[0027] The motor control instructions output by the collaborative control execution model after inputting the historical operation data of the target pump set are counted, and the cumulative distribution function of the instruction value is calculated;

[0028] The instruction value corresponding to the 95% quantile in the cumulative distribution function is selected as the pump set level control threshold;

[0029] If the motor control instruction currently running in the target pump set is greater than the pump set level control threshold, a control adjustment prompt is triggered, otherwise it is marked as normal operation.

[0030] Preferably, the specific method for periodically updating the parameters of the multi-motor collaborative calibration model is as follows:

[0031] Every fixed time period, collect the newly added operation data and working condition data of each target pump set, and perform format specification and interference elimination processing again to generate an updated standardized operation data set;

[0032] Use the updated standardized operation data set to retrain the multi-motor collaborative calibration model, adjust the parameters of the identification discriminator and the state calibrator through incremental learning, and maintain the calibration effect of the distribution of the motor state mapping vector and the working condition mapping vector.

[0033] Preferably, the processing method for the real-time data of the target pump set is as follows:

[0034] Divide the real-time data of the target pump set into time windows, extract the mean, variance, and maximum value in each time window as time series features, and perform dimension concatenation on the time series features and the multi-motor collaborative unified representation to obtain model input data.

[0035] Preferably, the construction method of the state mapping layer is as follows:

[0036] For numerical fields in the standardized operation data set, use normalization transformation to map them to the [0-1] interval; for text fields, use a keyword extraction algorithm to extract a text feature vector; concatenate the numerical mapping results and the text feature vector as input data for the state mapping layer, and output fixed-dimension motor state mapping vectors and working condition mapping vectors through a linear transformation layer.

[0037] Preferably, the specific process for triggering control adjustment is as follows:

[0038] When the motor control instruction currently running in the target pump set is greater than the pump set level control threshold, generate adjustment information containing the running time, pump set identifier, and control instruction;

[0039] The adjustment information is sent to the corresponding pump group management terminal, and is recorded to a control log database, and the log content includes adjustment time, adjustment level and associated operation data.

[0040] Compared with the prior art, the application has the beneficial effects that:

[0041] The method solves the problems of chaotic data format and large noise interference in traditional control by collecting historical operation data and real-time working condition data, and performing format specification and interference elimination processing to construct a standardized operation data set. Through unifying data units, completing missing values and filtering abnormal fluctuation points, a high-quality data foundation is provided for subsequent state modeling and collaborative control, ensuring the accuracy and consistency of model input, and reducing control deviation caused by data errors.

[0042] The construction of the multi-motor state mapping module realizes the mapping of motor state features and working condition features to a unified state space, converting structured data into quantifiable feature vectors. This mapping method not only retains the key information of motor operating states (such as speed, voltage) and working condition parameters (such as temperature, pressure), but also realizes the fusion representation of different types of features through a fixed-dimensional state space, making it possible to compare and collaboratively analyze the states of multiple motors, overcoming the defects of traditional methods in which feature dimensions are not unified and difficult to fuse.

[0043] The multi-motor collaborative calibration model effectively eliminates the state distribution differences between different motors through the collaborative optimization of the identification discriminator and the state calibrator, generating a multi-motor collaborative unified representation. This process ensures that the originally individual motor states overlap in the state space, ensuring that the collaborative control strategy can make decisions based on consistent global state awareness, avoiding uneven load distribution caused by individual motor differences, and improving the overall operation coordination of the pump group.

[0044] The collaborative control execution model takes the multi-motor collaborative unified representation and real-time data as input to generate precise motor control instructions, and combines a dynamic judgment mechanism of pump group level control thresholds to realize real-time adjustment and abnormal warning of control instructions. The 95% quantile threshold determined by historical data statistics can ensure stable operation under normal working conditions and timely trigger adjustments when the instructions exceed the reasonable range, reducing equipment wear and tear caused by abnormal control instructions and improving system safety and reliability.

[0045] Regularly updating the parameters of the multi-motor collaborative calibration model ensures that the control method can adapt to changes in working conditions and equipment aging over a long period of operation. Through incremental learning, the multi-motor collaborative calibration model is continuously optimized, maintaining the adaptability and robustness of the control strategy, prolonging the effective service life of the equipment, and reducing long-term maintenance costs. Attached Figure Description

[0046] Figure 1 A flowchart of a multi-motor coordinated control method for an intelligent multiphase booster pump device provided in this application;

[0047] Figure 2 The flowchart for generating the standardized runtime dataset provided in this application;

[0048] Figure 3 The flowchart for constructing the multi-motor state mapping module provided in this application;

[0049] Figure 4 The flowchart for constructing the multi-motor collaborative calibration model provided in this application;

[0050] Figure 5 The flowchart shows the training collaborative control execution model and the generation of motor control instructions provided in this application. Detailed Implementation

[0051] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0052] Please see Figures 1-5 This application provides a multi-motor coordinated control method for an intelligent multiphase booster pump device, and the specific implementation steps are as follows:

[0053] Historical and real-time operating data of multiphase booster pump units were collected. Both types of data underwent format standardization and interference removal to obtain a standardized operating dataset. From the monitoring systems of three types of pump groups—fluid transport, pressure regulation, and flow distribution—time stamps, motor speed, and drive voltage fields were extracted from historical operating data, while media temperature, pipeline pressure, and valve opening fields were extracted from real-time operating data. The extracted fields were then standardized in units and missing values ​​were filled in. Anomalies in the time stamps were filtered using a sliding window method. The processed historical and real-time operating data were stored separately as structured forms and then merged to obtain the standardized operating dataset.

[0054] A multi-motor state mapping module is constructed to map the motor state features and working condition features in the standardized running data set to the preset state space to generate motor state mapping vectors and working condition mapping vectors. A preset state model is selected as a basic mapping model, a state mapping layer is added to the input layer of the basic mapping model, for the numerical fields in the standardized running data set, a normalization transformation is used to map them to the [0-1] interval; for the text fields, a keyword extraction algorithm is used to extract the text feature vector; the numerical mapping result and the text feature vector are spliced as the input data of the state mapping layer, and the fixed-dimension motor state mapping vectors and working condition mapping vectors are output through the linear transformation layer. A state extractor is set in the middle layer of the basic mapping model, the output of the middle layer is mapped to the fixed-dimension state space through the linear transformation layer, and the motor state features and working condition features are encoded respectively to obtain the motor state mapping vectors and working condition mapping vectors corresponding to the pump group.

[0055] A multi-motor collaborative calibration model is constructed based on the motor state mapping vectors and working condition mapping vectors, and the distribution difference of different motor state mapping vectors is calibrated to obtain a multi-motor collaborative unified representation. Each motor is assigned a motor identifier, and the motor state mapping vector and the working condition mapping vector are taken as input to construct a collaborative network including an identifier discriminator and a state calibrator; the identifier discriminator is used to determine the motor identifier to which the input motor state mapping vector or working condition mapping vector belongs, and the state calibrator is used to adjust the distribution of the motor state mapping vector or working condition mapping vector, so that the motor state mapping vectors or working condition mapping vectors of different motor identifiers are distributed overlapping in the state space; by minimizing the classification error of the identifier discriminator and maximizing the calibration error of the state calibrator, the parameters of the multi-motor collaborative calibration model are optimized to obtain a multi-motor collaborative unified representation.

[0056] The collaborative control execution model is trained, and the multi-motor collaborative unified representation and the real-time data of the target pump group are used as input data, the motor control instruction is used as output data, and the motor control instruction currently running of the target pump group is generated. Collecting historical operation data and real-time working condition data of the target pump group, extracting the multi-motor collaborative unified representation thereof, marking the historical operation data as a control adjustment label, marking the real-time working condition data as a normal operation label, and constructing a training sample set; the time sequence features of the multi-motor collaborative unified representation and the real-time data of the target pump group are spliced as inputs, and a binary classification loss function is used to train the collaborative control execution model, so as to accurately predict the control adjustment label as a training target, and the model training is completed when the loss function converges; for the currently running target pump group, the multi-motor collaborative unified representation and real-time time sequence features thereof are extracted as input data, and the motor control instruction currently running of the target pump group is output. The processing method of the real-time data of the target pump group is: dividing the real-time data of the target pump group into time windows, extracting the mean, variance and maximum value in each time window as time sequence features, splicing the time sequence features and the multi-motor collaborative unified representation in dimension, and obtaining the model input data.

[0057] According to the instruction distribution of the historical operation data of the target pump group, the pump group level control threshold is calculated, the motor control instruction currently running of the target pump group is combined with the pump group level control threshold, it is judged whether the control adjustment is triggered, and the parameters of the multi-motor collaborative calibration model are updated regularly. The motor control instruction output by the collaborative control execution model after inputting the historical operation data of the target pump group is counted, and the cumulative distribution function of the instruction value is calculated; the instruction value corresponding to the 95% quantile in the cumulative distribution function is selected as the pump group level control threshold; if the motor control instruction currently running of the target pump group is greater than the pump group level control threshold, the control adjustment prompt is triggered, otherwise it is marked as normal operation. When the motor control instruction currently running of the target pump group is greater than the pump group level control threshold, adjustment information containing running time, pump group identifier and control instruction is generated; the adjustment information is sent to the corresponding pump group management terminal, and is recorded to the control log database at the same time, and the log content includes adjustment time, adjustment level and associated running data. Every fixed time period, collect the newly added running data and working condition data of each target pump group, reformat and remove interference, and generate an updated standardized running data set; the multi-motor collaborative calibration model is retrained using the updated standardized running data set, the parameters of the identification discriminator and the state calibrator are adjusted through incremental learning, and the calibration effect of the distribution of the different motor state mapping vectors and the working condition mapping vectors is maintained.

[0058] Embodiment 1:

[0059] In the process of constructing the multi-motor state mapping module, a preset state model needs to be selected as the basic mapping model, which can be determined in advance according to the operating characteristics and control requirements of the pump group. On this basis, a state mapping layer is added to the input layer of the basic mapping model. For numerical fields in the standardized operating data set, such as motor speed, drive voltage, medium temperature, and pipeline pressure, normalization transformation is used for processing. Through a linear transformation formula, the values of these fields are mapped to the interval range of [0-1]. This operation can standardize numerical fields, allowing numerical fields of different magnitudes to be compared and processed on the same dimension.

[0060] For text fields such as pump group types, since they are not in numerical form, they cannot be directly operated and processed numerically. Therefore, a keyword extraction algorithm is needed. This algorithm can extract representative text feature vectors from text fields, which can effectively represent the semantics and content of text fields. In operation, the text fields are first segmented, and irrelevant words such as stop words are removed. Then, from the remaining words, keywords that accurately describe the text content are extracted, and these keywords are converted into feature vector form.

[0061] After completing the normalization of numerical fields and the extraction of text feature vectors of text fields, the two results are concatenated in order to form a new vector as the input data of the state mapping layer. After receiving the input data, the state mapping layer processes it through a linear transformation layer (composed of a linear transformation matrix, with the row dimension consistent with the input data dimension and the column dimension fixed as a preset dimension). The input data is multiplied by the linear transformation matrix to convert it into a fixed-dimension state mapping vector and a working condition mapping vector. The fixed dimension is determined according to the requirements of the preset state model and the actual application scenario (such as 64 dimensions, 128 dimensions, etc., according to the number of pump motors and the dimension of working condition parameters), to ensure the uniformity of the output state mapping vector dimension and facilitate subsequent processing.

[0062] A state extractor is set in the middle layer of the basic mapping model, which maps the output of the middle layer to a fixed-dimension state space with the help of a linear transformation layer (also implemented by a linear transformation matrix, with the matrix parameters determined through model training optimization). Motor state features (such as motor speed and drive voltage) and working condition features (such as medium temperature, pipeline pressure, and valve opening) are encoded and converted into vectors in the state space to describe the motor operating state and the current working condition.

[0063] Through the above processing, the motor state mapping vector and the working condition mapping vector corresponding to the pump group are obtained. These mapping vectors map the motor state features and working condition features in the standardized running data set to the preset state space, which are used for subsequent construction of the multi-motor collaborative calibration model and implementation of the entire control method. During the processing, attention should be paid to the integrity and accuracy of the data, such as ensuring the correct range of numerical fields, reflecting the text content through keyword extraction, and reasonably setting the linear transformation matrix parameters to make the output state mapping vector and working condition mapping vector have good distinguishability and representativeness.

[0064] Embodiment 2:

[0065] When constructing the multi-motor collaborative calibration model based on the motor state mapping vector and the working condition mapping vector, a unique identification label (i.e., motor identification) is assigned to each motor. The motor identification can use numbers, letters, or specific symbol combinations to distinguish different motors and facilitate accurate identification of the state and features of the motor later.

[0066] The motor state mapping vector and the working condition mapping vector generated by the multi-motor state mapping module are used as inputs to construct a collaborative network containing an identification discriminator and a state calibrator. The collaborative network calibrates the distribution differences of different motor state mapping vectors through the collaborative work of the two. The identification discriminator can use a fully connected neural network structure (such as containing 2 layers of hidden layers, 32 neurons in each layer, and ReLU activation function), and the state calibrator can use a similar network structure (such as containing 1 layer of hidden layers, 16 neurons, and Sigmoid activation function).

[0067] The function of the identification discriminator is to judge the motor identification to which the input motor state mapping vector or working condition mapping vector belongs. It is essentially a classifier that learns the characteristics and rules of different mapping vectors through a large number of motor state mapping vectors or working condition mapping vectors and corresponding motor identification data. In actual work, when a new mapping vector is input, the identification discriminator predicts its corresponding motor identification based on the learned knowledge, and needs to be continuously trained and optimized to improve the judgment accuracy.

[0068] The state calibrator is responsible for adjusting the distribution of the motor state mapping vector or the working condition mapping vector. The core goal is to make the distributions of the motor state mapping vectors or working condition mapping vectors of different motor identifications as overlapping as possible in the state space. Due to factors such as manufacturing process differences and installation position deviations, the distributions of the motor state mapping vectors or working condition mapping vectors of different motors in the state space may differ. The state calibrator adjusts these mapping vectors through a specific algorithm to reduce the distribution differences and make the motor state mapping vectors or working condition mapping vectors of different motors more concentrated in the state space.

[0069] When training the collaborative control execution model, the performance of the identification discriminator and the state calibrator needs to be optimized simultaneously: by minimizing the classification error of the identification discriminator (i.e., the difference between the predicted motor identification and the actual motor identification, which can be calculated using a cross-entropy loss function), the parameters of the identification discriminator are adjusted to improve the judgment accuracy; by maximizing the calibration error of the state calibrator (reflecting the adjustment effect on the distribution of the mapping vector, such as using mean square error for reverse adjustment), it is prompted to more effectively adjust the distribution, so that the motor state mapping vectors of different motors and the working condition mapping vectors tend to be consistent.

[0070] The above two objectives have an antagonistic relationship: the identification discriminator needs to accurately distinguish the mapping vectors of different motors, while the state calibrator needs to make the distribution of these mapping vectors overlap and difficult to distinguish. Through adversarial training, the model gradually finds a balance point in iterations, so that the identification discriminator can distinguish the motor identification to some extent, while the state calibrator can effectively calibrate the distribution of the mapping vector.

[0071] When optimizing the parameters of the multi-motor collaborative calibration model, the classification error of the identification discriminator and the calibration error of the state calibrator are considered comprehensively, and the gradient descent algorithm is used to adjust and update the parameters. In each iteration, the parameter gradient is calculated according to the current error, and the parameters are updated in the opposite direction of the parameter gradient, gradually reducing the classification error and maximizing the calibration error.

[0072] After multiple iterations of training, when the error of the collaborative control execution model reaches a stable state or meets the preset convergence condition, the parameter optimization of the model is completed. At this time, the collaborative control execution model can effectively process the motor state mapping vectors and working condition mapping vectors of different motors, and obtain a unified representation of multi-motor collaboration, which eliminates the distribution differences of different motors and enables the state and characteristics of each motor to be represented and processed uniformly in the same state space.

[0073] In the process of building and training the multi-motor collaborative calibration model, it is necessary to ensure that the input motor state mapping vectors and working condition mapping vectors accurately reflect the actual running state and working condition, and the motor identification is accurately assigned, to avoid affecting the training and judgment of the identification discriminator. At the same time, the collaborative network structure (such as the number of network layers and the number of neurons of the identification discriminator and the state calibrator) is reasonably designed, and the learning rate and other hyperparameters are set to ensure stable training of the multi-motor collaborative calibration model and avoid overfitting or underfitting.

[0074] Embodiment 3:

[0075] In training the collaborative control execution model, historical operation data and real-time working condition data of the target pump group are collected. The historical operation data contains the running state information of the target pump group at different time points in the past, such as time mark, motor speed, driving voltage, etc. The real-time working condition data reflects the working environment and conditions of the current target pump group, including medium temperature, pipeline pressure, valve opening, etc. From these data, a unified representation of multi-motor collaboration is extracted, which is processed by the multi-motor collaborative calibration model to eliminate the distribution difference of different motor mapping vectors.

[0076] The collected data is labeled: the historical operation data is labeled as a control adjustment label, which indicates whether the motor control needs to be adjusted in the historical operation; the real-time working condition data is labeled as a normal operation label, which indicates whether the current working condition is normal, thereby constructing a training sample set to provide supervised learning data for the training of the multi-motor collaborative calibration model.

[0077] When processing the real-time data of the target pump group, the real-time data is divided into time windows, and the window size is determined according to the actual application scenario and data characteristics (such as 1 minute, 5 minutes). For the real-time data in each time window, the mean, variance, and maximum are extracted as time series features: the mean reflects the average level of real-time data, the variance reflects the dispersion degree of real-time data, and the maximum represents the maximum value in that time period, which describes the distribution and change of real-time data from different angles.

[0078] The extracted time series features are dimensionally spliced with the multi-motor collaborative unified representation. Specifically, the vector of time series features and the vector of multi-motor collaborative unified representation are sequentially connected to form a high-dimensional vector as the input data of the multi-motor collaborative calibration model. In this way, the information of time series features and multi-motor collaborative unified representation is fused together.

[0079] The collaborative control execution model is trained using a binary classification loss function, which is used to measure the difference between the predicted control adjustment label and the actual control adjustment label of the collaborative control execution model. Since the collaborative control execution model needs to predict whether the input data needs control adjustment, it belongs to a binary classification problem, and the training goal is to accurately predict the control adjustment label. In training, the input data is input into the collaborative control execution model to obtain the prediction result, and the loss value is calculated according to the difference between the prediction result (i.e. the predicted control adjustment label) and the actual control adjustment label. Through the back propagation algorithm, the loss value is transmitted to the parameters of the collaborative control execution model and the parameter gradient is calculated, and then the parameters are updated according to the parameter gradient to reduce the loss value. This process is repeated until the loss function converges, at which point the collaborative control execution model has learned the rules and characteristics of the data and can better predict the input data, and the training is completed.

[0080] According to the current operation of the target pump group, the multi-motor collaborative unified representation and real-time data thereof are extracted as input data: the multi-motor collaborative unified representation reflecting the collaborative state of each motor is obtained under the current running state, and the mean, variance and maximum of each time window are extracted as time series features by dividing the time window of the real-time data of the target pump group. After the time series features and the multi-motor collaborative unified representation are dimensionally spliced, the model input data of the collaborative control execution model are obtained.

[0081] After the trained collaborative control execution model receives the input data, it processes and analyzes the data according to the learned knowledge and rules, and outputs the motor control instructions of the target pump group under the current running state, such as motor speed adjustment value and drive voltage adjustment value, which are used to guide the operation of the motors in the target pump group and realize the control of the intelligent multi-phase booster pump device.

[0082] During the entire training and application process, attention should be paid to the quality and integrity of the data. The collected data must accurately reflect the actual running state and working condition of the target pump group, otherwise it will affect the training effect and prediction accuracy of the collaborative control execution model. When dividing the time window, the size of the time window should be reasonably set, otherwise it may lead to the extracted time series features not being able to well reflect the change rule of the data.

[0083] When labeling the data, it is necessary to ensure the accuracy of the control adjustment label and the normal running label, which directly affects the quality of the training sample set and further affects the training effect of the collaborative control execution model. When training the collaborative control execution model, the setting of hyperparameters such as learning rate and batch size is also very important, which needs to be reasonably adjusted according to the actual situation to ensure the stable training of the collaborative control execution model and avoid overfitting or underfitting.

[0084] When processing the real-time data of the target pump group, it is necessary to ensure the real-time and accuracy of the data, to obtain the latest running data and working condition data in time, and to perform corresponding feature extraction and processing, so as to ensure that the model input data input into the collaborative control execution model can accurately reflect the current running state, so that the motor control instructions output by the collaborative control execution model are targeted and effective.

[0085] Through the above detailed steps and processing process, the training of the collaborative control execution model and the function of generating the motor control instructions of the target pump group under the current running state by using the model are realized.

[0086] Example 4:

[0087] In the calculation of the pump group level control threshold according to the instruction distribution of the target pump group historical operation data, and in combination with the current running motor control instruction to judge whether to trigger control adjustment, first, the historical operation data of the target pump group is input into the collaborative control execution model to obtain the motor control instruction output by the collaborative control execution model. These instructions are specific control parameters generated by the collaborative control execution model for motor control under historical operation state, reflecting the control strategy of different historical scenarios.

[0088] Next, statistical analysis is performed on these motor control instructions to calculate the cumulative distribution function (CDF) of the motor control instructions. The cumulative distribution function is used to describe the probability of a random variable being less than or equal to a certain value, in this scenario, it can reflect the distribution probability of instruction values in different value intervals. Let the set of instruction values be , where represents the th instruction value, and the cumulative distribution function is defined as:

[0089] ,

[0090] In the formula, is the total number of instruction values in the historical operation data, is an indicator function, which is when the condition in the parentheses is true, otherwise. This formula constructs the cumulative distribution function by counting the proportion of instruction values less than or equal to a given value out of the total number of instruction values, thus clearly showing the distribution pattern of instruction values.

[0091] After obtaining the cumulative distribution function, the instruction value corresponding to the 95% quantile is selected as the pump group level control threshold. This quantile means that 95% of the historical motor control instructions are less than or equal to the pump group level control threshold, and only 5% of the instruction values are greater than the pump group level control threshold. Its selection is based on statistical analysis of historical operation data, and is used to determine the control limit for a small number of extreme conditions.

[0092] When the target pump group is currently running, its motor control instruction is obtained, and the instruction is compared with the pump group level control threshold . If , control adjustment is triggered; if , it is marked as normal operation. This judgment logic is based on the distribution characteristics of historical operation data, compares the current control instruction with the threshold of extreme cases in history to determine whether adjustment measures need to be taken.

[0093] When the control adjustment is triggered, adjustment information containing the runtime, pump group identification, and control instruction needs to be generated. The runtime is used to record the specific time when the control adjustment occurs, the pump group identification is used to specify which pump group needs to be adjusted, and the control instruction is the specific parameter that triggers the adjustment. These information constitutes the complete adjustment information, which can provide clear guidance for subsequent control operations.

[0094] After generating the adjustment information, it is sent to the corresponding pump group management terminal, and relevant personnel can understand the pump group that needs to be adjusted and the specific situation in time through the terminal, so as to take prompt measures. At the same time, the adjustment information is recorded to the control log database, and the log content includes the adjustment time, adjustment level, and associated running data, which provides information for subsequent analysis and tracing. The adjustment time needs to be consistent with the running time, the adjustment level is determined according to the instruction threshold value, which is used to distinguish the urgency and importance, and the associated running data includes medium temperature, pipeline pressure and other parameters related to the current adjustment.

[0095] During the process, the accuracy and completeness of data collection and processing need to be ensured. The quality of historical running data directly affects the calculation of cumulative distribution function and the determination of pump group level control threshold, and the true and reliable historical running data needs to be ensured to avoid unreasonable pump group level control threshold setting due to errors. The collection of current running data needs to be timely and accurate to ensure that the motor control instruction obtained can truly reflect the current state of the target pump group and avoid misjudgment affecting the adjustment opportunity. When calculating the cumulative distribution function, the selection of 95% quantile is based on common engineering practice and statistical method, and in actual application, it can be adjusted according to the operation requirements and risk bearing capacity of the target pump group. In high-risk scenarios, 90% quantile can be selected to meet the control requirements of different scenarios.

[0096] When recording the adjustment information to the control log database, the integrity and traceability of the log content need to be ensured. Detailed log recording not only helps subsequent analysis and summary of control adjustment events, but also provides data support for optimization and improvement of the system. Through analysis of historical adjustment logs, common problems and potential risks in the operation of the target pump group can be found, so that preventive measures can be taken to improve the stability and reliability of the system.

[0097] Through the above steps, the pump group level control threshold is determined based on the instruction distribution of the historical running data, and whether the adjustment is triggered is judged by combining the motor control instruction of the current running of the target pump group. The process strictly follows the principle of data-driven, realizes the accurate monitoring and timely adjustment of the running state of the target pump group through statistical analysis of historical running data and real-time comparison of current data, and ensures that the intelligent multi-phase booster pump device can run in a safe and stable state.

[0098] Example 5:

[0099] When updating the parameters of the multi-motor collaborative calibration model periodically, a fixed time period is first determined. This period is set according to the actual operation of the intelligent multi-phase booster pump device, and the fluid delivery pump group with stable working conditions can be set to every week, and the pressure regulating pump group with variable working conditions can be set to every day. When the period is reached, the newly added operation data and working condition data of each pump group are collected.

[0100] Taking a fluid delivery pump group as an example, during the operation in the past week, its monitoring system will continuously record new historical operation data such as time mark, motor speed, and driving voltage, and also real-time collect operation data and working condition data such as medium temperature, pipeline pressure, and valve opening. These newly added operation data and working condition data are the basis for updating the multi-motor collaborative calibration model, and need to be collected comprehensively and accurately to ensure that important operation information, including sudden data under abnormal working conditions, is not missed.

[0101] After collecting the newly added operation data and working condition data of each target pump group, these data need to be formatted and interference removed. Specifically, for the time mark, motor speed, and driving voltage fields in the newly added operation data of the fluid delivery pump group, and the medium temperature, pipeline pressure, and valve opening fields in the newly added working condition data, unit uniformity processing is required first. For example, motor speed may be recorded in "turns / minute" or "rpm", which needs to be converted to "turns / minute"; medium temperature may be represented in Celsius or Fahrenheit, which needs to be converted to Celsius using the conversion formula .

[0102] Missing values in the operation data and working condition data are completed. For example, during a certain period of time, due to sensor failure, the pipeline pressure data of the fluid delivery pump group is missing, which needs to be completed using appropriate methods. The missing pressure value can be estimated using linear interpolation method (t2 time value = 0.4 + (0.6 - 0.4) x (t2 - t1) / (t3 - t1)) based on the pipeline pressure data at the previous and next time points (e.g., t1 time is 0.4 MPa, t3 time is 0.6 MPa, t2 time is missing), or using the average value of the pipeline pressure data under similar working conditions of the same pump group.

[0103] Abnormal fluctuation points in the time mark are filtered using the sliding window method. For example, the motor speed data of the fluid delivery pump group suddenly fluctuates greatly at a certain time point, which obviously deviates from the normal operating range, which may be caused by external interference or data collection error. By setting a suitable sliding window size (e.g., 10-minute window), the mean and standard deviation of the data within the time window are calculated, and the data points that exceed a certain multiple of the standard deviation of the mean are considered as abnormal points and are filtered or corrected.

[0104] After the above processing, the processed operating data and the working condition data are respectively stored as structured forms. The structured forms include a table form, a row-column corresponding record, and fields including "time mark", "motor speed", "drive voltage", etc. Then, the structured forms are merged to generate an updated standardized operating data set.

[0105] After generating the updated standardized operating data set, the multi-motor collaborative calibration model needs to be retrained using the updated standardized operating data set. Taking the three types of pump groups, i.e., the fluid delivery pump group, the pressure regulating pump group, and the flow distribution pump group, as examples, the motor state mapping vector and the working condition mapping vector in the updated standardized operating data set are input into the multi-motor collaborative calibration model.

[0106] In the training process, the parameters of the identification discriminator and the state calibrator are adjusted in an incremental learning manner. Incremental learning refers to further learning and optimization using the updated standardized operating data set on the basis of retaining the knowledge learned by the previous multi-motor collaborative calibration model. For example, the previous multi-motor collaborative calibration model can well process the motor state mapping vector of the fluid delivery pump group under the conventional working condition, but as new data are added, a new type of high-temperature working condition may appear. At this time, through incremental learning, the multi-motor collaborative calibration model can learn how to process these new working conditions without forgetting the previously learned knowledge.

[0107] For the identification discriminator, its role is to judge the motor identification to which the input motor state mapping vector or working condition mapping vector belongs. In the incremental learning process, the network parameters of the identification discriminator are adjusted according to the input data, so that it can more accurately identify the motor identification of different motors in the updated standardized operating data set. For example, when a new motor type appears in the updated standardized operating data set, the identification discriminator needs to learn to accurately identify the motor state mapping vector or working condition mapping vector of the motor.

[0108] For the state calibrator, its role is to adjust the distribution of the motor state mapping vector or the working condition mapping vector, so that the mapping vectors of different motor identifications overlap in the state space. In the incremental learning process, the parameters of the state calibrator are adjusted according to the distribution of the mapping vectors of different motors in the updated standardized operating data set, to maintain the calibration effect on the distribution difference of the mapping vectors of different motors in the updated standardized operating data set. For example, when the distribution of the motor state mapping vector or the working condition mapping vector of the fluid delivery pump group and the pressure regulating pump group in the updated standardized operating data set appears a new difference, the state calibrator needs to adjust the parameters so that the distribution of the motor state mapping vector or the working condition mapping vector of the newly added data can be effectively calibrated.

[0109] By continuously performing incremental learning, adjusting the parameters of the identification discriminator and the state calibrator, the multi-motor collaborative calibration model can adapt to the characteristics of the new data and maintain the calibration effect of the mapping vectors of different motor states and the mapping vectors of working conditions. In this way, over time, the multi-motor collaborative calibration model can continuously learn new knowledge and improve its calibration ability for multi-motor collaborative states, thereby better adapting to the needs of the intelligent multi-phase supercharged pump device in different operating stages.

[0110] During the entire periodic updating process, attention needs to be paid to the timeliness and accuracy of data collection. If the newly collected operating data and working condition data are not timely, it may cause the multi-motor collaborative calibration model to fail to adapt to changes in the pump set operating state in a timely manner; if the newly collected operating data and working condition data are not accurate, it will affect the training effect of the multi-motor collaborative calibration model, resulting in deviations in the calibration of the motor state mapping vectors and the working condition mapping vectors by the multi-motor collaborative calibration model.

[0111] The parameter adjustment process of incremental learning needs to be reasonably controlled in terms of learning rate. If the learning rate is too large, it may cause the multi-motor collaborative calibration model to forget previously learned knowledge when learning new data; if the learning rate is too small, the learning speed of the multi-motor collaborative calibration model for new data will be slow, and it will not be able to adapt to changes in data in a timely manner.

[0112] Through the above detailed steps and processing procedures, periodic updating of the parameters of the multi-motor collaborative calibration model is achieved. Taking the newly generated data of fluid delivery pump sets, pressure regulating pump sets, and flow distribution pump sets in actual operation as an example, by periodically collecting and processing these data and retraining the multi-motor collaborative calibration model using incremental learning, the multi-motor collaborative calibration model can continuously maintain the calibration effect of the mapping vectors of different motor states and the mapping vectors of working conditions, thereby ensuring that the multi-motor collaborative control method of the intelligent multi-phase supercharged pump device can always maintain good performance and adapt to changing operating conditions.

[0113] It should be noted that in this document, relational terms such as first and second and the like can be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0114] While embodiments of the application have been shown and described, it is to be understood that the embodiments described are merely exemplary and that changes can be made in detail without departing from the principles and spirit of the application. The scope of the application is therefore defined by the appended claims and their equivalents.

Claims

1. A method for controlling multi-motor coordination for intelligent multiphase booster pump apparatus, characterized by, The application relates to a method for realizing multi-motor collaborative control of a multi-phase booster pump device. The method comprises the following steps: collecting historical operation data and real-time working condition data of the multi-phase booster pump device, performing format specification and interference elimination processing on the two types of data respectively, and obtaining a standardized operation data set; constructing a multi-motor state mapping module, mapping motor state features and working condition features in the standardized operation data set to a preset state space, and generating a motor state mapping vector and a working condition mapping vector; constructing a multi-motor collaborative calibration model based on the motor state mapping vector and the working condition mapping vector, calibrating the distribution differences of different motor mapping vectors, and obtaining a multi-motor collaborative unified representation; training a collaborative control execution model, taking the multi-motor collaborative unified representation and target pump group real-time data as input data, and taking motor control instructions as output data, and generating motor control instructions for the current operation of the target pump group; 2. The method for control of multi motor coordination for intelligent multiphase booster pump arrangement as claimed in claim 1 wherein, calculating a pump group level control threshold according to the instruction distribution of the historical operation data of the target pump group, combining the motor control instructions for the current operation of the target pump group and the pump group level control threshold, judging whether to trigger control adjustment, and regularly updating the parameters of the multi-motor collaborative calibration model. The specific method for collecting the historical operation data and real-time working condition data of the multi-phase booster pump device, performing format specification and interference elimination processing on the two types of data respectively, and obtaining a standardized operation data set is as follows: extracting time markers, motor speeds and driving voltages in the historical operation data and medium temperatures, pipeline pressures and valve opening degrees in the real-time working condition data from monitoring systems of three types of pump groups, namely fluid conveying, pressure regulating and flow distribution; 3. The method for multi-motor coordinated control of an intelligent multiphase booster pump device according to claim 2, characterized in that, performing unit unification and missing value completion processing on the extracted fields, filtering abnormal fluctuation points in the time markers by using a sliding window method, storing the processed historical operation data and real-time working condition data into structured forms respectively, and merging to obtain a standardized operation data set. The specific method for constructing a multi-motor state mapping module, mapping motor state features and working condition features in the standardized operation data set to a preset state space, and generating a motor state mapping vector is as follows: selecting a preset state model as a basic mapping model, adding a state mapping layer to the input layer of the basic mapping model, converting the structured fields in the standardized operation data set into feature vector forms, setting a state extractor in the middle layer of the basic mapping model, mapping the output of the middle layer to a fixed-dimensional state space through a linear transformation layer, and respectively encoding the motor state features and the working condition features to obtain motor state mapping vectors and working condition mapping vectors under corresponding pump groups.

4. The method for multi-motor coordinated control of an intelligent multiphase booster pump device of claim 3, wherein, The specific method for constructing a multi-motor collaborative calibration model based on the motor state mapping vector and the working condition mapping vector, calibrating the distribution differences of different motor mapping vectors, and obtaining a multi-motor collaborative unified representation is as follows: allocating motor identifiers to each motor, taking the motor state mapping vector and the working condition mapping vector as input, and constructing a collaborative network comprising an identifier discriminator and a state calibrator; the identifier discriminator is used for judging the motor identifier to which the input motor state mapping vector or working condition mapping vector belongs, and the state calibrator is used for adjusting the distribution of the motor state mapping vector or working condition mapping vector, so that the motor state mapping vectors or working condition mapping vectors of different motor identifiers are distributed in the state space and overlap. The parameters of the multi-motor collaborative calibration model are optimized by minimizing the classification error of the identification discriminator and maximizing the calibration error of the state calibrator, and a multi-motor collaborative unified representation is obtained.

5. The method for multi-motor coordinated control of an intelligent multiphase booster pump device of claim 4, wherein, The specific method for generating the motor control instruction currently running in the target pump group is that the collaborative control execution model is trained with the multi-motor collaborative unified representation and real-time data of the target pump group as input data and with the motor control instruction as output data. The historical operation data and real-time working condition data of the target pump group are collected, the multi-motor collaborative unified representation is extracted, the historical operation data is labeled as a control adjustment label, the real-time working condition data is labeled as a normal operation label, and a training sample set is constructed; The time sequence features of the multi-motor collaborative unified representation and the real-time data of the target pump group are spliced as input, and the collaborative control execution model is trained using a binary classification loss function to accurately predict the control adjustment label as the training target, and the model training is completed when the loss function converges; For the currently running target pump group, the multi-motor collaborative unified representation and real-time time sequence features are extracted as input data, and the motor control instruction currently running in the target pump group is output.

6. The method for multi-motor coordinated control of a smart multiphase booster pump arrangement of claim 5, wherein, The specific method for calculating the pump group level control threshold value according to the instruction distribution of the historical operation data of the target pump group, combining the motor control instruction currently running in the target pump group with the pump group level control threshold value, and judging whether to trigger control adjustment is as follows: The cumulative distribution function of the instruction value is calculated by statistically analyzing the motor control instruction output by the collaborative control execution model after inputting the historical operation data of the target pump group; The instruction value corresponding to the 95% quantile in the cumulative distribution function is selected as the pump group level control threshold value; If the motor control instruction currently running in the target pump group is greater than the pump group level control threshold value, control adjustment is triggered, otherwise it is marked as normal operation.

7. The method for the control of multi-motor coordination for intelligent multiphase booster pump arrangement of claim 6, wherein, The specific method for periodically updating the parameters of the multi-motor collaborative calibration model is as follows: Every fixed time period, collect the newly added operation data and working condition data of each target pump group, reformat and remove interference, and generate an updated standardized operation data set; The multi-motor collaborative calibration model is retrained using the updated standardized operation data set, and the parameters of the identification discriminator and the state calibrator are adjusted through incremental learning to maintain the calibration effect of the distribution of the motor state mapping vector and the working condition mapping vector.

8. The method for multi-motor coordinated control of an intelligent multiphase booster pump device of claim 5, wherein, The processing method of the real-time data of the target pump group is as follows: The real-time data of the target pump group is divided into time windows, the mean, variance, and maximum values in each time window are extracted as time sequence features, and the time sequence features and the multi-motor collaborative unified representation are dimensionally spliced to obtain model input data.

9. The method for multi-motor coordinated control of an intelligent multiphase booster pump device of claim 3, wherein, The construction method of the state mapping layer is as follows: For numerical fields in the standardized operation data set, use normalization transformation to map them to the [0-1] interval; for text fields, use keyword extraction algorithm to extract text feature vectors; splice the numerical mapping results and the text feature vectors as input data of the state mapping layer, and output fixed-dimension motor state mapping vectors and working condition mapping vectors through a linear transformation layer.

10. The method for multi-motor coordinated control of an intelligent multiphase booster pump device of claim 6, wherein, The specific process of triggering control adjustment is as follows: When the current motor control instruction of the target pump group is greater than the pump group level control threshold, adjustment information including the running time, pump group identification, and control instruction is generated; The adjustment information is sent to the corresponding pump group management terminal, and is recorded to the control log database, and the log content includes the adjustment time, adjustment level, and associated running data.

Images