Compressor mass flow calculation method

Abstract

The invention discloses a compressor mass flow calculation method, which is based on a long short-term memory neural network, and comprises the following steps of: (1) selecting characteristic parameters related to mass flow measurement from historical data by using a mutual information method to carry out standardized preprocessing; (2) constructing the characteristic parameters into a multi-parameter multi-step time sequence data sample according to a time sequence; (3) inputting the data sample into a long short-term memory neural network model for training and optimization, and obtaining a quality flow prediction value; and (4) comparing a mass flow predicted value with a measured value, and adjusting hyper-parameters of the long and short-term memory neural network model until a predicted result meets a precision requirement. On the premise that hardware cost is not increased, the mass flow of the compressor can be rapidly and accurately calculated, and test time and money cost are greatly reduced.

Description

Technical Field

[0001] The present invention relates to the field of air conditioning, and in particular to a method for calculating the mass flow of a compressor. Background Art

[0002] Compressors, the driving force behind the refrigerant circulation in compression refrigeration systems, are core components of refrigeration equipment. Compressor performance testing is an essential step for compressor manufacturers in numerous areas, including product development and improvement, trial runs, batch production inspections, and quality checks. Among the various parameters tested for compressor performance, refrigerant mass flow directly reflects compressor capacity and is a critical parameter that must be measured. Improving the speed and reducing testing costs of compressor refrigerant mass flow testing are key to improving the efficiency of large-scale compressor testing.

[0003] Currently, compressor manufacturers use methods recommended in the national standard "GB-T5773-2016 Performance Test Method for Positive Displacement Refrigerant Compressors" to calculate refrigerant mass flow. The second refrigerant calorimeter method recommended in the national standard does not require any additional components to be installed in the refrigeration system. Its measurement method closely resembles the actual operating conditions of the refrigeration system, making it the primary test method for refrigerant mass flow measurement. The second refrigerant calorimeter method works by continuously adjusting the electrical heating power of the calorimeter to equal the cooling capacity of the refrigeration system, until the system reaches thermal equilibrium. The mass flow measurement is then derived from the ratio of the electrical heating power to the enthalpy difference between the inlet and outlet of the evaporator coil. However, due to the large heat capacity of the calorimeter itself, the refrigeration system typically requires over an hour to reach thermal equilibrium. This is especially true when the refrigerant flow rate of the compressor being tested is low, requiring even longer stabilization time. Therefore, directly using the second refrigerant calorimeter method specified in the national standard cannot meet the demand for rapid compressor mass flow measurement.

[0004] Existing optimizations for shortening test time using the second refrigerant calorimeter method are all achieved by adding additional components. For example, the published paper "Optimization Design of Compressor Test Equipment for the Second Refrigerant Calorimeter Method (2017)" proposes that the addition of a high-pressure regulating valve and a small electronic expansion valve can quickly respond to the compressor exhaust pressure, thereby shortening the operating condition stabilization time. The published paper "PLC Fuzzy Control of the Structure and Heating Capacity of the Second Refrigerant Calorimeter (2014)" uses PLC combined with PID control technology to achieve stepless automatic control of the electric heating power, allowing the suction temperature to stabilize quickly. These improvements require the addition of additional components to the test system, significantly increasing the cost and complexity of the test system. Summary of the Invention

[0005] The Summary of the Invention introduces a series of simplified concepts, which are simplifications of existing technologies in the field and are further described in detail in the Detailed Description of the Invention. The Summary of the Invention is not intended to define the key features and essential features of the claimed technical solution, nor is it intended to determine the scope of protection of the claimed technical solution.

[0006] The technical problem to be solved by the present invention is to provide a method for quickly and accurately calculating the mass flow measurement of a compressor without increasing the hardware cost.

[0007] To solve the above technical problems, the present invention provides a method for calculating compressor mass flow rate, which is based on a long short-term memory neural network and includes the following steps:

[0008] (1) Using the mutual information method to select characteristic parameters related to mass flow measurement from historical data for standardization preprocessing;

[0009] Mutual Information Method (MIM) is a statistical method based on information theory that measures the dependency between two random variables. Mutual information can be used to assess the correlation between two variables, not just the linear correlation, and is therefore widely used in many fields (such as machine learning, signal processing, and bioinformatics).

[0010] (2) Constructing the characteristic parameters into multi-parameter multi-step time series data samples according to the time series;

[0011] (3) Inputting the data samples into the long short-term memory neural network model for training and optimization to obtain the mass flow prediction value;

[0012] Adaptive Moment Estimation (Adam) is an optimization algorithm for parameter optimization, widely used in deep learning and machine learning. Adam combines the advantages of momentum and adaptive learning rate, and can effectively handle sparse gradients and non-stationary targets.

[0013] (4) Compare the predicted mass flow rate with the measured value and adjust the hyperparameters of the long short-term memory neural network model until the prediction result meets the accuracy requirements.

[0014] Preferably, the compressor mass flow calculation method is further improved by selecting the following characteristic parameters related to mass flow measurement;

[0015] Firstly, time, calorimeter inlet and outlet temperature and pressure, and calorimeter heating power are selected as characteristic parameters;

[0016] Then, the mutual information method is used to select the five features with the highest mutual information values with mass flow as input features.

[0017] Preferably, the compressor mass flow calculation method is further improved, and the standardization preprocessing includes: filtering processing and normalization processing.

[0018] Preferably, the compressor mass flow calculation method is further improved, and the filtering process uses the Savitzky-Golay filtering method to smooth the data and filter out high-frequency data generated by system vibration in the original data;

[0019] The selected feature parameters are normalized using the minimum and maximum normalization method so that the normalized results are mapped to within [0,1].

[0020] Preferably, the compressor mass flow calculation method is further improved, and the constructed multi-parameter multi-step time series data samples are in matrix form, each row includes historical data before a certain moment and future data after that moment, and the data samples are arranged in order according to the time steps from row to row, and optimized using the adaptive moment estimation algorithm for parameter optimization.

[0021] Preferably, the compressor mass flow calculation method is further improved, and when implementing step (3), the historical data column in the data sample is used as the input independent variable X, and the future data column in the data sample is used as the input supervision variable y.

[0022] Preferably, the compressor mass flow calculation method is further improved, and when implementing step (3), the long short-term memory neural network layer introduces three judgment conditions of input gate, forget gate and output gate to the neurons;

[0023] The input gate represents the ratio of information allowed to be added to the memory cell;

[0024] The forget gate indicates the proportion of historical information stored in the current state node;

[0025] The output gate indicates the ratio of the information of the current state node to be output.

[0026] Preferably, the compressor mass flow calculation method is further improved, and the activation function of the dense layer of the long short-term memory neural network layer is selected as the Sigmoid function.

[0027] Preferably, the compressor mass flow calculation method is further improved, and during the long short-term memory neural network model training process, the mean square error is selected as the target loss function, and the adaptive moment estimation algorithm is selected as the gradient descent optimization algorithm.

[0028] Preferably, the compressor mass flow calculation method is further improved, and the normalized data is used as the data to be predicted, and is input into the trained long short-term memory neural network layer to calculate the normalized prediction result, and the result is denormalized and compared with the measured result. If the prediction error is large, the model hyperparameters are adjusted until the result meets the accuracy requirements.

[0029] The main design concept of this invention is to propose a prediction model to partially replace compressor mass flow testing, thereby shortening the total test duration. The specific principle is based on a long-short-term memory neural network prediction model. By learning historical test data over a period of time, the learned algorithm model predicts the mass flow rate after the system stabilizes. This allows the flow rate value in the stable phase to be obtained without having to conduct the test to the stable phase.

[0030] Compared with the prior art, the present invention can at least achieve the following technical effects:

[0031] 1. The present invention adopts a long short-term memory neural network, which can selectively use historical operating data of several parameters within 30 minutes, including calorimeter heating power, outlet pressure, expansion valve inlet temperature and pressure, to predict the mass flow rate after the compressor performance test bench stabilizes. There is no need to wait until the test bench stabilizes, which greatly reduces the time and money costs of the test.

[0032] 2. This paper utilizes the Adaptive Moment Estimation (Adam) algorithm to optimize the traditional gradient descent algorithm. By estimating the expected value of the gradient, it avoids random walks in the gradient itself and reduces the risk of the model converging to a local optimum. Furthermore, by setting an upper limit on the learning step size for each iteration, the Adaptive Moment Estimation algorithm ensures that the parameter values during each iteration are relatively stable, thus avoiding the problem of exploding gradients.

[0033] 3. This method filters the data before training because during the test, mechanical vibrations or fluctuations in the internal refrigerant can cause high-frequency fluctuations in the data collected by the system. This can lead to large differences in data values at adjacent acquisition time points, hindering the determination of data stability and model fitting and prediction. The Savitzky-Golay filtering method effectively removes high-frequency data while preserving the data's inherent trend characteristics.

[0034] 4. The present invention normalizes the data before training because the range of experimental data is large and the measurement units and sizes of the collected data are inconsistent. Therefore, normalizing the data effectively reduces the impact of initial values that are too large or too small, so that all input data are mapped within a range that is easy to calculate, thereby making network training more efficient.

[0035] 5. The present invention reconstructs the data samples so that each data input to the model contains historical data before a certain moment and future data after that moment, which matches the input parameter requirements of LSTM and enables the model to learn the implicit relationship between historical data and future data. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The drawings herein are intended to illustrate the general characteristics of methods, structures, and / or materials used in certain exemplary embodiments of the present invention, supplementing the descriptions in the specification. However, the drawings herein are schematic diagrams not drawn to scale and may not accurately reflect the precise structure or performance characteristics of any given embodiment. The drawings herein should not be interpreted as defining or limiting the range of values or properties encompassed by the exemplary embodiments of the present invention. The present invention is further described in detail below in conjunction with the drawings and specific embodiments:

[0037] Figure 1 It is a schematic flow diagram of the present invention.

[0038] Figure 2 3 is a schematic diagram comparing the mass flow rate data of the present invention before and after using Savitzky-Golay filtering.

[0039] Figure 3 It is a schematic diagram of the long short-term memory neural network model constructed by the present invention.

[0040] Figure 4 Schematic diagram of the model training error curve constructed by the present invention.

[0041] Figures 5 to 7 It is a schematic diagram comparing the predicted results of the model test set constructed by the present invention with the measured results. DETAILED DESCRIPTION

[0042] The following describes the embodiments of the present invention through specific embodiments. Those skilled in the art will fully understand the other advantages and technical effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through different specific embodiments, and the details in this specification can be applied based on different perspectives and various modifications or changes can be made without departing from the overall design concept of the invention. It should be noted that the following embodiments and features therein can be combined with each other unless there is a conflict. The following exemplary embodiments of the present invention can be implemented in a variety of different forms and should not be construed as being limited to the specific embodiments described herein. It should be understood that these embodiments are provided to make the disclosure of the present invention thorough and complete and to fully convey the technical solutions of these exemplary embodiments to those skilled in the art. It should be understood that when an element is referred to as being "connected" or "coupled" to another element, the element can be directly connected or coupled to the other element, or intervening elements can be present. In contrast, when an element is referred to as being "directly connected" or "directly coupled" to another element, there are no intervening elements. Throughout the drawings, the same reference numerals represent the same element. As used herein, the term "and / or" includes any and all combinations of one or more of the relevant listed items.

[0043] Example;

[0044] The present invention provides a method for calculating compressor mass flow rate, which is based on a long short-term memory neural network and includes the following steps:

[0045] (1) The specific selection is to select some relevant feature parameters through domain knowledge, and then select the features most relevant to mass flow through the mutual information method; mutual information is a measure of the degree of dependence between two random variables. The larger the mutual information value, the higher the degree of dependence between the two; the mutual information calculation function can be expressed as:

[0046]

[0047] For example, based on professional knowledge, we first selected the time, calorimeter heating power, expansion valve inlet temperature and pressure, and calorimeter outlet temperature and pressure data collected from a compressor performance test bench. Then, we used Equation (1) to calculate the mutual information value between each feature and the mass flow data. The calculated results were 1.17, 1.34, 1.18, 0.52, 1.23, and 1.29, respectively. Among them, the mutual information value between the calorimeter outlet temperature and the mass flow rate was only 0.52, which was significantly lower than the results obtained for other parameters, indicating that the dependence between the calorimeter outlet temperature parameter and the mass flow rate was not high, so it was eliminated.

[0048] Standardization preprocessing includes: filtering and normalization;

[0049] The Savitzky-Golay filtering method is used to smooth the data and filter out the high-frequency data generated by system vibration in the original data. Savitzky-Golay can effectively remove high-frequency data and retain the data's own change trend characteristics. The Savitzky-Golay filtering formula is:

[0050]

[0051] The selected data is smoothed using the Savitzky-Golay filter of formula (2), and the filter window length is preferably set to 65 and the polynomial order is 1. Taking mass flow data as an example, the effects before and after filtering are as follows: Figure 2 As shown. Using the min-max normalization method in Equation (3), the results are mapped to the range [0, 1]. Specifically, all training data must be normalized, and the normalization dimension is set for each column of input data. This is because each column of data represents an indicator, and normalization is meaningful only for data within the range of this indicator.

[0052] The selected feature parameters are normalized using the minimum and maximum normalization method so that the normalized results are mapped to [0, 1]. The purpose of normalization is to eliminate the impact of the dimension difference of the input data on the prediction results. The function of the minimum and maximum normalization method can be expressed as:

[0053]

[0054] (2) Constructing the characteristic parameters into multi-parameter multi-step time series data samples according to the time series, specifically including:

[0055] The historical data of the feature parameters before a certain moment and the future data after that moment are constructed in the same data row. The next row is the data row of the next moment. The reconstructed data sample is in matrix form. To further illustrate the feature parameter reconstruction process, the following is an example of a data with two feature parameters:

[0056] The data sample construction process is as follows: All data before time t in the original data is a d×2 matrix D, where var1(t) is the data of independent variable 1 recorded at time t, and var2(t) is the data of independent variable 2 recorded at time t. The m data before time t [D(tm), ..., D(t-2), D(t-1)] are used as the input independent variable X, and the n data at time t and after [D(t), D(t+1), ..., D(t+n-1)] are used as the input supervision y. The reconstructed input independent variable X is a (dm-n+1)×2m matrix, and the input supervision y is a (dm-n+1)×2n matrix.

[0057]

[0058] The data is reconstructed according to equations (4) to (6), where the number of historical data items is 60 and the number of future data items is 60, i.e., m = 60 and n = 60. The normalized data is a matrix with 844 rows and 6 columns, and the reconstructed data is a matrix with 725 rows and 720 columns. The first 360 columns of the matrix are the model input X, and the last 360 columns are the model input y.

[0059] (3) Inputting the data samples into the long short-term memory neural network model for training and optimization to obtain the mass flow prediction value;

[0060] The reconstructed data samples X and y are input into a long-short-term memory neural network model to determine the optimal combination of hyperparameters, including the network step size, neuron deactivation coefficient, and batch size. The network step size is a structural parameter of the LSTM neural network, representing the process from input to output and back to input during the neural network operation. As the step size increases, the neural network's operational accuracy improves, but the operation time also increases. An appropriate neuron deactivation coefficient and batch size can prevent the model from overfitting. The neuron deactivation coefficient represents the probability of discarding a portion of neurons during the neural network operation. By applying this parameter, the regularization method originally required in the neural network is replaced, avoiding overfitting. An appropriate batch size can avoid the simultaneous calculation of all sample data, saving computation time.

[0061] The reconstructed data is divided into training set and test set in the ratio of 3:1. The training set is input into the model. The sigmoid function is selected as the activation function, the GlorotUniform method is selected as the weight initialization method, the mean square error is selected as the target loss function, and the adaptive moment estimation algorithm is selected as the gradient descent optimization algorithm. The LSTM neural network model is set to have 2 hidden layers. After grid optimization, the network step size of the first layer is determined to be 50, the network step size of the second layer is determined to be 20, the neuron deactivation coefficient is determined to be 0.2, and the batch size is determined to be 64. The constructed LSTM neural network structure is as follows: Figure 3 shown. Figure 4 This is an error curve diagram of the model constructed according to an embodiment of the present invention in the training test. The loss after 50 iterations is already small, so the epoch of model training is set to 50.

[0062] (4) The normalized data is used as the data to be predicted and is input into the trained long short-term memory neural network layer to calculate the normalized prediction results. The test set X is input into the model, and the predicted value given by the model is denormalized and compared with the measured results. If the prediction error is large, the model hyperparameters are adjusted until the result meets the accuracy requirements. The results within [0,1] are mapped back to the original data size to facilitate direct comparison with the experimental test values. The denormalization function of the minimum and maximum normalization method can be expressed as:

[0063] X′ pre =X pre (X max -X min )+X min (7)

[0064] The comparison between the predicted and measured mass flow rates is shown in Figure 2. Figure 5 As shown in Figure 2, the maximum relative error of the predicted value occurs at the end of the test set, but the relative error of the terminal prediction value is still only 0.68%, and the prediction result is accurate and reliable. The prediction results of the other two groups of data are compared with the measured results. Figure 6 、 Figure 7 As shown, the maximum error is less than 1%.

[0065] For ease of description, spatially relative terms such as "below", "above", "below", "above", "above", etc. may be used herein to describe the spatial positional relationship of an element or feature to other elements or features as shown in the figures. It should be understood that spatially relative terms are intended to encompass different orientations of the device in use or operation in addition to the orientation depicted in the figures. For example, if the device in the drawings is inverted, the element described as "below other elements or features" or "below other elements or features" will be subsequently positioned as "above other elements or features" or "above other elements or features". Thus, the exemplary term "below" may include both "above" and "below". The device may also be positioned in other different ways (rotated 90 degrees or in other orientations), and the spatially relative descriptors used herein are interpreted accordingly.

[0066] The terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should also be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of the features, wholes, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.

[0067] Example embodiments according to the present invention are described herein with reference to schematic cross-sectional views of preferred embodiments (and intermediate structures) as exemplary embodiments. As such, variations in the shapes shown, for example, due to manufacturing techniques and / or tolerances, are expected. Therefore, the exemplary embodiments should not be interpreted as being limited to the specific shapes of the regions shown herein, but may also include deviations in shape, for example, due to manufacturing. For example, an implanted region shown as a rectangle may have rounded or curved features and / or a gradient of implant concentration at its edges, rather than just a binary change from an implanted region to a non-implanted region. Similarly, a buried region formed by implantation may result in some implantation in the region between the buried region and the surface through which the implantation is passed. Therefore, the regions shown in the figures are schematic in nature, and their shapes are not intended to illustrate the actual shapes of the regions in the device, nor are they intended to limit the scope of the example embodiments according to the present invention.

[0068] Unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. It will also be understood that, unless expressly defined herein, terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant art, rather than being interpreted in an idealized or overly formal sense.

[0069] The present invention has been described in detail above by way of specific embodiments and examples, but these do not constitute limitations of the present invention. Without departing from the principles of the present invention, those skilled in the art may make many variations and improvements, which should also be considered within the scope of protection of the present invention.

Claims

1. A method for calculating compressor mass flow rate based on a long short-term memory neural network, characterized in that: The following steps are involved: (1) Using the mutual information method to select characteristic parameters related to mass flow measurement from historical data for standardization preprocessing; (2) Constructing the characteristic parameters into multi-parameter multi-step time series data samples according to the time series; (3) Inputting the data samples into the long short-term memory neural network model for training and optimization to obtain the mass flow prediction value; (4) Compare the predicted mass flow rate with the measured value and adjust the hyperparameters of the long short-term memory neural network model until the prediction result meets the accuracy requirements.

2. The method for calculating the mass flow of a compressor according to claim 1, wherein: The following characteristic parameters related to mass flow measurement are selected; Firstly, time, calorimeter inlet and outlet temperature and pressure, and calorimeter heating power are selected as characteristic parameters; Then, the mutual information method is used to select the five features with the highest mutual information values with mass flow as input features.

3. The compressor mass flow calculation method according to claim 1, wherein: Standardization preprocessing includes filtering and normalization.

4. The method for calculating the mass flow of a compressor according to claim 3, wherein: Filtering: The Savitzky-Golay filtering method is used to smooth the data and filter out high-frequency data generated by system vibration in the original data; The selected feature parameters are normalized using the minimum and maximum normalization method so that the normalized results are mapped to within [0,1].

5. The method for calculating the mass flow of a compressor according to claim 1, wherein: The constructed multi-parameter and multi-step time series data samples are in matrix form. Each row includes historical data before a certain moment and future data after that moment. The data samples are arranged row by row according to the time step and optimized using the adaptive moment estimation algorithm for parameter optimization.

6. The method for calculating the mass flow of a compressor according to claim 5, wherein: When implementing step (3), the historical data column in the data sample is used as the input independent variable X, and the future data column in the data sample is used as the input supervision variable y.

7. The method for calculating the mass flow of a compressor according to claim 6, wherein: When implementing step (3), the long short-term memory neural network layer introduces three judgment conditions for neurons: input gate, forget gate, and output gate; The input gate represents the ratio of information allowed to be added to the memory cell; The forget gate indicates the proportion of historical information stored in the current state node; The output gate indicates the ratio of the information of the current state node to be output.

8. The method for calculating the mass flow of a compressor according to claim 7, wherein: The Sigmoid function is selected as the activation function of the dense layer of the long short-term memory neural network layer.

9. The method for calculating the mass flow of a compressor according to claim 1, wherein: During the training process of the long short-term memory neural network model, the mean square error is selected as the target loss function, and the adaptive moment estimation algorithm is selected as the gradient descent optimization algorithm.

10. The method for calculating the mass flow of a compressor according to claim 5, wherein: The normalized data is used as the data to be predicted and input into the trained long short-term memory neural network layer to calculate the normalized prediction results. The results are denormalized and compared with the measured results. If the prediction error is large, the model hyperparameters are adjusted until the results meet the accuracy requirements.

Images