A compressor dynamic protection control method and electronic starting controller

Abstract

The application provides a compressor dynamic protection control method and an electronic starting controller, through scene tagging of operation data of the compressor, various load scenes are divided, and a basic critical value offset coefficient is set; parameters of the compressor are collected in real time, real-time load intensity is calculated through a weighting algorithm, and a buffer interval is determined; data of the compressor are acquired and converted into a time sequence characteristic sequence, after normalization processing, the time sequence characteristic sequence is then input into a neural network model of a pre-trained double-branch network structure, and a trend level and a PID parameter interval are output; according to the trend level and the PID parameter interval, basic PID parameters are determined, a risk deviation coefficient is introduced, and final PID parameters are determined; an output quantity of the final PID parameters is set with a smooth transition limiting amplitude, and the compressor is operated under the smooth transition limiting amplitude, so that the delay and curve fluctuation problems of traditional critical value control are solved.

Description

Technical Field

[0001] This invention belongs to the field of compressor dynamic protection and control technology, and specifically relates to a compressor dynamic protection and control method and an electronic starter. Background Technology

[0002] A compressor is a driven fluid machine that raises low-pressure gas to high-pressure gas. It draws in low-temperature, low-pressure refrigerant gas through the suction pipe, compresses it, and then discharges high-temperature, high-pressure refrigerant gas through the discharge pipe, providing power for the refrigeration cycle.

[0003] Traditional compressors have a threshold value set for over-temperature or over-current protection. If the temperature or current exceeds the threshold value, the compressor control is activated. However, this control has a certain control delay, which means that there will still be brief over-temperature or over-current events. In addition, the control curve of the entire compressor will fluctuate. Summary of the Invention

[0004] Based on this, the present invention provides a compressor dynamic protection control method and electronic starter, which aims to solve the problems of delay and curve fluctuation in traditional critical value control, and realize early intervention and smooth regulation of the compressor.

[0005] The first aspect of this invention provides a dynamic protection control method for a compressor, which tags the compressor's operating data into various load scenarios and sets a basic critical value offset coefficient for each load scenario.

[0006] The compressor parameters are collected in real time, and the real-time load intensity is calculated using a weighted algorithm. The parameters include at least the operating current, winding temperature, and refrigerant pressure.

[0007] The buffer zone is determined based on the traditional critical value, the load intensity, and the basic critical value offset coefficient;

[0008] The compressor data is acquired and converted into a time-series feature sequence. The data in the time-series feature sequence is normalized based on the buffer space. Then, it is input into a pre-trained neural network model with a dual-branch network structure. The output is the trend level and PID parameter range. The compressor data includes the operating current, the winding temperature, the refrigerant pressure, the outdoor ambient temperature, the indoor set temperature, the fan speed, the rate of change of the winding temperature within a preset time, and the rate of change of the operating current within a preset time. The neural network model with a dual-branch network structure includes a trend prediction branch and a PID parameter generation branch. The trend prediction branch is an LSTM layer + Attention network, and the PID parameter generation branch is a fully connected layer + residual network network. In addition, when data falls into the buffer space, the Attention mechanism assigns higher weights to the corresponding data.

[0009] Based on the trend level and the PID parameter range, the basic PID parameters are determined, and a risk deviation coefficient is introduced. Based on the basic PID parameters and the risk deviation coefficient, the final PID parameters are determined.

[0010] Set a smooth transition limit for the final PID parameter output, and run the compressor under the smooth transition limit.

[0011] Furthermore, in the step of real-time acquisition of compressor parameters and calculation of real-time load intensity using a weighted algorithm, the weights corresponding to the acquired compressor parameters are obtained through training with historical fault data.

[0012] Furthermore, the step of obtaining the weights corresponding to the parameters of the collected compressor through training with historical fault data includes:

[0013] Obtain historical operating data of the compressor, and extract current characteristics, temperature characteristics, pressure characteristics, and tags from the historical operating data;

[0014] After data cleaning and enhancement of the extracted current features, temperature features, pressure features, and labels, logistic regression is used as the weight learning model. During the training process, gradient descent is used to optimize the weights and determine the final weights.

[0015] Furthermore, in the step of determining the buffer zone based on the traditional critical value, the load intensity, and the basic critical value offset coefficient, the calculation formula for the buffer zone is as follows:

[0016] The upper limit between buffer zones = traditional critical value × (1 - basic critical value offset coefficient × S);

[0017] The lower limit between buffer zones = traditional critical value × (1 - basic critical value offset coefficient × S - 0.1);

[0018] Wherein, S is the load intensity.

[0019] Furthermore, the steps of acquiring compressor data, converting it into a time-series feature sequence, normalizing the data in the time-series feature sequence based on the buffer interval, and then inputting it into a pre-trained neural network model with a dual-branch network structure to output trend levels and PID parameter ranges include:

[0020] The compressor data is acquired and converted into a time-series feature sequence. The data in the time-series feature sequence is normalized based on the buffer space, and then input into the LSTM layer to output a global time-series feature vector.

[0021] Based on the global time series feature vector and the buffer, risk-sensitive time series features are determined;

[0022] Based on the aforementioned risk-sensitive time-series characteristics, output the trend level;

[0023] The trend level is concatenated with the working condition features and used as the input to the fully connected layer, in which a residual network structure is added to prevent overfitting of parameter mapping.

[0024] The results of the residual network are input into the output layer, and the PID parameters are mapped to the target range based on the Sigmoid activation.

[0025] Furthermore, the step of determining the basic PID parameters based on the trend level and the PID parameter range, introducing a risk deviation coefficient, and determining the final PID parameters based on the basic PID parameters and the risk deviation coefficient includes:

[0026] Based on the trend level and the PID parameter range, determine the basic PID parameters;

[0027] Based on the risk deviation coefficient, the risk urgency of the current basic PID parameters corresponding to the position within the buffer zone is quantified, and the control intensity of the basic PID parameters is dynamically amplified / reduced to determine the final PID parameters.

[0028] Furthermore, in the step of quantifying the risk urgency of the current basic PID parameters corresponding to the location within the buffer zone based on the risk deviation coefficient, the formula for calculating the risk urgency is as follows:

[0029] Risk deviation coefficient = (upper limit between buffer zones - actual value) / (upper limit between buffer zones - lower limit between buffer zones);

[0030] Corrected deviation = Traditional deviation × (2 - Risk deviation coefficient).

[0031] A second aspect of this invention provides an electronic starter controller for implementing the compressor dynamic protection control method described in the first aspect, the electronic starter controller comprising:

[0032] The segmentation module is used to tag the compressor's operating data into different scenarios, classify them into various load scenarios, and set basic critical value offset coefficients for each type of load scenario.

[0033] The calculation module is used to collect compressor parameters in real time and calculate the real-time load intensity through a weighted algorithm. The parameters include at least the operating current, winding temperature and refrigerant pressure.

[0034] The first determining module is used to determine the buffer zone based on the traditional critical value, the load intensity, and the basic critical value offset coefficient.

[0035] The input module is used to acquire compressor data and convert it into a time-series feature sequence. The data in the time-series feature sequence is normalized based on the buffer space. Then, it is input into a pre-trained neural network model with a dual-branch network structure, which outputs the trend level and PID parameter range. The compressor data includes the operating current, the winding temperature, the refrigerant pressure, the outdoor ambient temperature, the indoor set temperature, the fan speed, the rate of change of the winding temperature within a preset time, and the rate of change of the operating current within a preset time. The neural network model with a dual-branch network structure includes a trend prediction branch and a PID parameter generation branch. The trend prediction branch is an LSTM layer + Attention network, and the PID parameter generation branch is a fully connected layer + residual network network. In addition, when data falls into the buffer space, the Attention mechanism assigns higher weights to the corresponding data.

[0036] The second determining module is used to determine the basic PID parameters based on the trend level and the PID parameter range, introduce a risk deviation coefficient, and determine the final PID parameters based on the basic PID parameters and the risk deviation coefficient.

[0037] The setting module is used to set a smooth transition limit for the output of the final PID parameters, and to run the compressor under the smooth transition limit.

[0038] This invention provides a dynamic protection control method and electronic starter for a compressor. The method involves tagging the compressor's operating data to identify various load scenarios and setting a basic critical value offset coefficient for each scenario. It also involves real-time acquisition of compressor parameters and calculation of real-time load intensity using a weighted algorithm. These parameters include at least operating current, winding temperature, and refrigerant pressure. A buffer zone is determined based on the traditional critical value, load intensity, and the basic critical value offset coefficient. The compressor data is acquired and converted into a time-series feature sequence. The data in the time-series feature sequence is normalized using the buffer zone as a reference, and then input into a pre-trained dual-branch network. In the neural network model of the network structure, the output trend level and PID parameter range are generated. Based on the trend level and PID parameter range, the basic PID parameters are determined, and a risk deviation coefficient is introduced. Based on the basic PID parameters and the risk deviation coefficient, the final PID parameters are determined. A smooth transition limit is set for the output of the final PID parameters, and the compressor runs under the smooth transition limit. Specifically, it breaks through the traditional fixed critical value thinking and realizes the scenario adaptation between buffer zones through load intensity calculation, solving the problem of inaccurate pre-intervention timing. At the same time, it combines trend prediction with PID parameter generation, so that PID changes from passive response to active prediction, reducing control fluctuations from the root. Attached Figure Description

[0039] Figure 1This is a flowchart illustrating the implementation of a compressor dynamic protection control method according to Embodiment 1 of the present invention.

[0040] Figure 2 This is a structural block diagram of an electronic starter provided in Embodiment 2 of the present invention. Detailed Implementation

[0041] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0042] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0043] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0044] Example 1

[0045] According to an embodiment of the present invention, a dynamic protection control method for a compressor is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0046] This first embodiment provides a dynamic protection control method for a compressor. Please refer to [link / reference]. Figure 1 , Figure 1 The flowchart of a compressor dynamic protection control method provided in Embodiment 1 of the present invention is shown, specifically including steps S01 to S06.

[0047] Step S01: The compressor's operating data is tagged with scenarios to classify various load scenarios, and a basic critical value offset coefficient is set for each type of load scenario.

[0048] It should be noted that existing technologies use fixed threshold values ​​for protection and control, which cannot adapt to compressor load fluctuations (such as summer cooling peaks, winter heating start-up, voltage fluctuations, etc.). This results in the buffer zone being either too narrow (not playing a pre-intervention role) or too wide (wasting energy). Therefore, in the embodiments of this invention, the buffer zone is adapted to the specific scenario by calculating the load intensity, thus solving the problem of inaccurate pre-intervention timing.

[0049] Specifically, the operating data of the compressor throughout its entire life cycle is first tagged with scenarios, dividing it into 6 core load scenarios, including light load scenarios (such as low-load operation in spring and autumn, indoor and outdoor temperature difference <5℃), normal load scenarios (such as stable operation during the day in summer, temperature difference 5-10℃), heavy load scenarios (such as high-temperature start-up in the afternoon in summer, temperature difference >10℃), start-up impact scenarios (within 10 minutes after the compressor is powered on, instantaneous peak current), voltage fluctuation scenarios (when the grid voltage fluctuates by ±10%), and abnormal approach scenarios (such as the temperature slowly rising due to condenser blockage, approaching the traditional critical value). Each scenario corresponds to a set of basic critical value offset coefficients (such as offset coefficient of 0.2 for light load scenarios and offset coefficient of 0.3 for heavy load scenarios), avoiding the drawbacks of adapting all scenarios to a single buffer.

[0050] Step S02: Collect compressor parameters in real time and calculate real-time load intensity using a weighted algorithm. The parameters include at least operating current, winding temperature, and refrigerant pressure.

[0051] Specifically, the compressor parameters are collected in real time, including operating current I, winding temperature T, and refrigerant pressure P. The load intensity S is calculated by weighted summation. The weights of operating current I, winding temperature T, and refrigerant pressure P are obtained through training with historical fault data.

[0052] More specifically, historical operating data of the compressor is acquired, and current characteristics, temperature characteristics, pressure characteristics, and tags are extracted from the historical operating data. In this embodiment of the invention, the historical operating data of the compressor includes at least 3,000 complete operating cycles, including at least 500 cases of over-temperature / over-current faults, covering different operating conditions (such as high temperature in summer, low temperature in winter, voltage fluctuations, etc.) and equipment status (such as new machine, machine that has been running for 1 year, and machine that has been running for 3 years). It can be understood that the current characteristic refers to the average current value within 1 minute before the fault occurs, the temperature characteristic refers to the average winding temperature within 1 minute before the fault occurs, the pressure characteristic refers to the average refrigerant high pressure within 1 minute before the fault occurs, and the tags include 1 and 0. 1 indicates that the operating state corresponding to this set of characteristics ultimately led to the over-temperature / over-current fault, and 0 indicates that the operating state corresponding to this set of characteristics did not experience a fault (normal operation).

[0053] After data cleaning and enhancement of the extracted current, temperature, pressure features, and labels, logistic regression is used as the weight learning model. During training, the weights are optimized through gradient descent to determine the final weights. The output of logistic regression is the failure probability, which can directly reflect the correlation strength between features and failure risk. The model parameters (coefficients) can be directly used as the weights of each feature. In addition, the cross-entropy loss function is used to measure the model prediction error, with the goal of minimizing the loss.

[0054] Step S03: Determine the buffer zone based on the traditional critical value, the load intensity, and the basic critical value offset coefficient.

[0055] Specifically, the formula for calculating the buffer space is as follows:

[0056] The upper limit between buffer zones = traditional critical value × (1 - basic critical value offset coefficient × S);

[0057] The lower limit between buffer zones = traditional critical value × (1 - basic critical value offset coefficient × S - 0.1);

[0058] Wherein, S is the load intensity.

[0059] For example, in a heavy-load scenario, S=0.8, the heavy-load scenario offset coefficient=0.3, and the traditional critical temperature is 80℃. Then the buffer zone is 80×(1-0.30.8)~80×(1-0.3×0.8-0.1)=60.8℃~52.8℃. Pre-intervention is initiated 20℃ in advance to avoid the temperature "overshooting".

[0060] Step S04: Obtain the compressor data and convert it into a time-series feature sequence. Normalize the data in the time-series feature sequence based on the buffer interval, and then input it into a pre-trained neural network model with a dual-branch network structure to output the trend level and PID parameter range.

[0061] In this embodiment of the invention, the role of the dual-branch neural network model is to identify the parameter change trend within the buffer zone in advance and provide predictive parameters for PID control, avoiding PID control fluctuations caused by lag response. Specifically, the compressor data includes the operating current, winding temperature, refrigerant pressure, outdoor ambient temperature, indoor set temperature, fan speed, rate of change of winding temperature within a preset time, and rate of change of operating current within a preset time. The operating current, winding temperature, and refrigerant pressure are core risk features; the outdoor ambient temperature, indoor set temperature, and fan speed are operating condition features; and the rate of change of winding temperature and operating current within a preset time are historical trend features. The dual-branch neural network model includes a trend prediction branch and a PID parameter generation branch. The trend prediction branch is an LSTM layer + Attention network, and the PID parameter generation branch is a fully connected layer + residual network network. In addition, when data falls into the buffer zone, the Attention mechanism assigns higher weights to the corresponding data. It can be understood that among the features input to the neural network model, only when the parameters enter the buffer zone do these features have high-risk warning value. For example, when the winding temperature is far below the lower limit of the buffer zone, the small fluctuations in the rate of change of the winding temperature within a preset time are meaningless to the system safety, and the neural network will automatically reduce its weight; however, when the winding temperature enters the buffer zone, the fluctuations in the same rate of change of the winding temperature within a preset time will be given extremely high weight (reinforced by the Attention mechanism), ensuring that the model focuses on the scenarios that really need intervention.

[0062] It should be noted that the compressor data is acquired and converted into a time-series feature sequence. The data in this sequence is normalized using the buffer space as a reference, and then input into an LSTM layer to output a global time-series feature vector. The LSTM layer includes a first LSTM layer and a second LSTM layer. The first LSTM layer has 32 neurons and is used to learn local time-series patterns, employing the tanh activation function to preserve time-series direction information. The second LSTM layer has 16 neurons and is used to learn global trend features, incorporating Dropout (0.2) to suppress overfitting. Understandably, the neural network model normalizes the input features using the buffer space as a reference rather than traditional critical values. For example, the normalization formula for the winding temperature feature is:

[0063] T_norm = (Current winding temperature - Lower limit of buffer zone) / (Upper limit of buffer zone - Lower limit of buffer zone);

[0064] This allows the feature values ​​to accurately reflect the risk distance from the upper limit of the buffer zone within the [0,1] interval, rather than the absolute distance from the traditional critical value, making the model more sensitive to subtle changes within the buffer zone;

[0065] Based on the global time-series feature vector and the buffer, risk-sensitive time-series features are determined. Specifically, firstly, a risk query vector is constructed, consisting of the upper limit of the buffer, the current parameter value, and historical fault-related features. Then, the similarity between the global time-series feature vector and the risk query vector is calculated to obtain the attention weight of each feature. High-risk features (features falling into the buffer) are forcibly given a higher weight (multiplied by a factor of 1.5) to ensure that the model prioritizes critical risk signals. Finally, the global time-series feature vector and the attention weight are multiplied element by element to obtain the risk-sensitive time-series features.

[0066] Based on the risk-sensitive time series characteristics, a trend level is output. In this embodiment of the invention, there are 5 trend levels, including sudden drop, gradual drop, stable, gradual rise, and sudden rise.

[0067] The trend level and working condition features are concatenated and used as the input to the fully connected layer. A residual network structure is added to prevent overfitting of the parameter mapping. It should be noted that the residual network structure includes three residual blocks in series. Each residual block contains two fully connected layers and a skip connection. The first layer has 64 neurons (ReLU activation) and the second layer has 32 neurons (ReLU activation). The skip connection directly superimposes the input to the output (solving the gradient vanishing problem in deep networks).

[0068] The results of the residual network are input into the output layer. Based on the Sigmoid activation, the PID parameters are mapped to the target interval. In this embodiment of the invention, the Sigmoid interval mapping is as follows:

[0069] Kp = 0.1 + 0.9 × Sigmoid(Kp_raw) (mapped to [0.1, 1.0]);

[0070] Ki=0.01+0.09×Sigmoid(Ki_raw) (mapped to [0.01,0.1]);

[0071] Kd = 0.001 + 0.009 × Sigmoid(Kd_raw) (mapped to [0.001, 0.01]);

[0072] Kp is the proportional coefficient, Ki is the integral coefficient, Kd is the differential coefficient, Kp_raw is the original proportional coefficient, Ki_raw is the original integral coefficient, and Kd_raw is the original differential coefficient. The original proportional coefficient, original integral coefficient, and original differential coefficient are the results of the residual network.

[0073] In addition, the trend prediction branch of the neural network will judge in real time whether the current buffer zone is reasonable. For example, when the model predicts that "the winding temperature will rise sharply at a rate of 0.8℃ / s", and the current buffer zone only reserves a buffer space of 5℃ (theoretically, 5℃ / 0.8℃ / s=6.25s response time), but the actual PID control requires at least 8s to suppress the temperature rise, the model will indirectly promote the temporary expansion of the buffer zone (such as adding a 2℃ buffer) by outputting a high-risk trend level, so as to avoid the control failure caused by the preset range being too narrow.

[0074] Step S05: Determine the basic PID parameters based on the trend level and the PID parameter range, and introduce a risk deviation coefficient. Determine the final PID parameters based on the basic PID parameters and the risk deviation coefficient.

[0075] Specifically, based on the trend level and the PID parameter range, the basic PID parameters are determined. For example, if the trend is "sharp increase": assign a larger Kp (to enhance the proportional control strength and quickly suppress parameter increases), a smaller Ki (to avoid overshoot caused by integral accumulation), and a medium Kd (to weaken derivative oscillations); if the trend is "gradual increase": assign a medium Kp, a larger Ki (to slowly offset static deviations and avoid temperature fluctuations), and a smaller Kd (to reduce the sensitivity of the derivative to small changes); if the trend is "stable": assign a smaller Kp, a medium Ki, and a smaller Kd (to maintain stability and avoid over-control).

[0076] Based on the risk deviation coefficient, the risk urgency of the current basic PID parameters corresponding to the location within the buffer zone is quantified. The control intensity of the basic PID parameters is dynamically amplified / reduced to determine the final PID parameters. It is understandable that the trend type is also a gradual increase. The risk level is different when the distance from the upper limit of the buffer zone is 1℃ and when the distance from the upper limit is 5℃. Therefore, this step is to make the parameters accurately match the risk distance.

[0077] The formula for calculating the urgency of a quantitative risk is as follows:

[0078] Risk deviation coefficient = (upper limit between buffer zones - actual value) / (upper limit between buffer zones - lower limit between buffer zones);

[0079] Corrected deviation = Traditional deviation × (2 - Risk deviation coefficient).

[0080] For example, with an actual winding temperature of 60℃ and a buffer zone temperature of 52.8℃~60.8℃, the risk deviation is (60.8-60) / (60.8-52.8)=0.1, and the corrected deviation is (60.8-60)(2-0.1)=0.8+1.9=1.52, which is larger than the traditional deviation of 0.8, indicating stronger PID control and preventing the winding temperature from exceeding the upper limit of the buffer zone.

[0081] Step S06: Set a smooth transition limit for the output of the final PID parameters, and run the compressor under the smooth transition limit.

[0082] Specifically, the single control quantity change amplitude is less than or equal to the maximum allowable change, the control quantity change rate is less than or equal to the maximum change rate, and the PID output is processed by a first-order low-pass filter to further reduce the instantaneous fluctuations of the control quantity and ensure a smooth transition in the compressor's operating state.

[0083] In summary, the compressor dynamic protection control method in the above embodiments of the present invention involves: labeling the compressor's operating data to classify various load scenarios and setting basic critical value offset coefficients for each load scenario; collecting compressor parameters in real time and calculating the real-time load intensity using a weighted algorithm, with parameters including at least operating current, winding temperature, and refrigerant pressure; determining a buffer zone based on traditional critical values, load intensity, and basic critical value offset coefficients; acquiring compressor data and converting it into a time-series feature sequence; normalizing the data in the time-series feature sequence using the buffer zone as a reference; and then inputting the data into a pre-trained dual-branch network. In the neural network model of the structure, the output trend level and PID parameter range are output. Based on the trend level and PID parameter range, the basic PID parameters are determined, and a risk deviation coefficient is introduced. Based on the basic PID parameters and the risk deviation coefficient, the final PID parameters are determined. A smooth transition limit is set for the output of the final PID parameters, and the compressor runs under the smooth transition limit. Specifically, it breaks through the traditional fixed critical value thinking and realizes the scenario adaptation between buffer zones through load intensity calculation, solving the problem of inaccurate pre-intervention timing. At the same time, it combines trend prediction with PID parameter generation, so that PID changes from passive response to active prediction, reducing control fluctuations from the root.

[0084] Example 2

[0085] Please see Figure 2 , Figure 2 This is a structural block diagram of an electronic starter 200 provided in Embodiment 2 of the present invention. The electronic starter 200 is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that implements a predetermined function. Although the apparatus described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0086] Specifically, the electronic starter 200 includes: a division module 21, a calculation module 22, a first determination module 23, an input module 24, a second determination module 25, and a setting module 26, wherein:

[0087] The segmentation module 21 is used to tag the compressor's operating data into various load scenarios and set basic critical value offset coefficients for each load scenario.

[0088] The calculation module 22 is used to collect the parameters of the compressor in real time and calculate the real-time load intensity through a weighted algorithm. The parameters include at least the operating current, winding temperature and refrigerant pressure. The weights corresponding to the collected compressor parameters are obtained by training with historical fault data.

[0089] The first determining module 23 is used to determine the buffer zone based on the traditional critical value, the load intensity, and the basic critical value offset coefficient. The calculation formula for the buffer zone is as follows:

[0090] The upper limit between buffer zones = traditional critical value × (1 - basic critical value offset coefficient × S);

[0091] The lower limit between buffer zones = traditional critical value × (1 - basic critical value offset coefficient × S - 0.1);

[0092] Wherein, S is the load intensity;

[0093] Input module 24 is used to acquire compressor data and convert it into a time-series feature sequence. The data in the time-series feature sequence is normalized based on the buffer space. Then, it is input into a pre-trained neural network model with a dual-branch network structure, which outputs the trend level and PID parameter range. The compressor data includes the operating current, the winding temperature, the refrigerant pressure, the outdoor ambient temperature, the indoor set temperature, the fan speed, the rate of change of the winding temperature within a preset time, and the rate of change of the operating current within a preset time. The neural network model with a dual-branch network structure includes a trend prediction branch and a PID parameter generation branch. The trend prediction branch is an LSTM layer + Attention network, and the PID parameter generation branch is a fully connected layer + residual network network. In addition, when data falls into the buffer space, the Attention mechanism assigns higher weights to the corresponding data.

[0094] The second determining module 25 is used to determine the basic PID parameters based on the trend level and the PID parameter range, introduce a risk deviation coefficient, and determine the final PID parameters based on the basic PID parameters and the risk deviation coefficient.

[0095] Setting module 26 is used to set a smooth transition limit for the output of the final PID parameters, and to run the compressor under the smooth transition limit.

[0096] Furthermore, in some optional embodiments of the present invention, the computing module 22 includes:

[0097] The acquisition unit is used to acquire historical operating data of the compressor and extract current characteristics, temperature characteristics, pressure characteristics, and tags from the historical operating data;

[0098] The weight determination unit is used to clean and enhance the extracted current features, temperature features, pressure features, and labels, and then uses logistic regression as the weight learning model. During the training process, the weights are optimized through gradient descent to determine the final weights.

[0099] Furthermore, in some optional embodiments of the present invention, the input module 24 includes:

[0100] The normalization processing unit is used to acquire the compressor data and convert it into a time-series feature sequence. It normalizes the data in the time-series feature sequence based on the buffer space and then inputs it into the LSTM layer to output a global time-series feature vector.

[0101] The first determining unit is used to determine risk-sensitive time-series features based on the global time-series feature vector and the buffer space.

[0102] The output unit is used to output the trend level based on the risk-sensitive time series characteristics.

[0103] The splicing unit is used to splice the trend level with the working condition features as the input of the fully connected layer. A residual network structure is added to prevent overfitting of parameter mapping.

[0104] The mapping unit is used to input the results of the residual network into the output layer and map the PID parameters to the target range based on the Sigmoid activation.

[0105] Furthermore, in some optional embodiments of the present invention, the second determining module 25 includes:

[0106] The second determining unit is used to determine the basic PID parameters based on the trend level and the PID parameter range;

[0107] The control unit is used to quantify the risk urgency of the current base PID parameters corresponding to the position within the buffer zone based on the risk deviation coefficient, dynamically amplify / reduce the control intensity of the base PID parameters, and determine the final PID parameters. The formula for calculating the quantified risk urgency is as follows:

[0108] Risk deviation coefficient = (upper limit between buffer zones - actual value) / (upper limit between buffer zones - lower limit between buffer zones);

[0109] Corrected deviation = Traditional deviation × (2 - Risk deviation coefficient).

[0110] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0111] The above embodiments merely illustrate several implementation methods of the present invention, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this patent should be determined by the appended claims.

Claims

1. A dynamic protection control method for a compressor, characterized in that, The method includes: The compressor's operating data is tagged with scenarios to classify various load scenarios, and a basic critical value offset coefficient is set for each type of load scenario. The compressor parameters are collected in real time, and the real-time load intensity is calculated using a weighted algorithm. The parameters include at least the operating current, winding temperature, and refrigerant pressure. The buffer zone is determined based on the traditional critical value, the load intensity, and the basic critical value offset coefficient; The compressor data is acquired and converted into a time-series feature sequence. The data in the time-series feature sequence is normalized based on the buffer space. Then, it is input into a pre-trained neural network model with a dual-branch network structure. The output is the trend level and PID parameter range. The compressor data includes the operating current, the winding temperature, the refrigerant pressure, the outdoor ambient temperature, the indoor set temperature, the fan speed, the rate of change of the winding temperature within a preset time, and the rate of change of the operating current within a preset time. The neural network model with a dual-branch network structure includes a trend prediction branch and a PID parameter generation branch. The trend prediction branch is an LSTM layer + Attention network, and the PID parameter generation branch is a fully connected layer + residual network network. In addition, when data falls into the buffer space, the Attention mechanism assigns higher weights to the corresponding data. Based on the trend level and the PID parameter range, the basic PID parameters are determined, and a risk deviation coefficient is introduced. Based on the basic PID parameters and the risk deviation coefficient, the final PID parameters are determined. Set a smooth transition limit for the final PID parameter output, and run the compressor under the smooth transition limit.

2. The compressor dynamic protection control method according to claim 1, characterized in that, In the step of real-time acquisition of compressor parameters and calculation of real-time load intensity using a weighted algorithm, the weights corresponding to the acquired compressor parameters are obtained through training with historical fault data.

3. The compressor dynamic protection control method according to claim 2, characterized in that, The steps involved in obtaining the weights corresponding to the parameters of the compressor through training with historical fault data include: Obtain historical operating data of the compressor, and extract current characteristics, temperature characteristics, pressure characteristics, and tags from the historical operating data; After data cleaning and enhancement of the extracted current features, temperature features, pressure features, and labels, logistic regression is used as the weight learning model. During the training process, gradient descent is used to optimize the weights and determine the final weights.

4. The compressor dynamic protection control method according to claim 3, characterized in that, In the step of determining the buffer zone based on the traditional critical value, the load intensity, and the basic critical value offset coefficient, the calculation formula for the buffer zone is as follows: The upper limit between buffer zones = traditional critical value × (1 - basic critical value offset coefficient × S); The lower limit between buffer zones = traditional critical value × (1 - basic critical value offset coefficient × S - 0.1); Wherein, S is the load intensity.

5. The compressor dynamic protection control method according to claim 4, characterized in that, The steps of acquiring compressor data, converting it into a time-series feature sequence, normalizing the data in the time-series feature sequence based on the buffer interval, and then inputting it into a pre-trained neural network model with a dual-branch network structure to output trend levels and PID parameter ranges include: The compressor data is acquired and converted into a time-series feature sequence. The data in the time-series feature sequence is normalized based on the buffer space, and then input into the LSTM layer to output a global time-series feature vector. Based on the global time series feature vector and the buffer, risk-sensitive time series features are determined; Based on the aforementioned risk-sensitive time-series characteristics, output the trend level; The trend level is concatenated with the working condition features and used as the input to the fully connected layer, in which a residual network structure is added to prevent overfitting of parameter mapping. The results of the residual network are input into the output layer, and the PID parameters are mapped to the target range based on the Sigmoid activation.

6. The compressor dynamic protection control method according to claim 5, characterized in that, The steps of determining the basic PID parameters based on the trend level and the PID parameter range, introducing a risk deviation coefficient, and determining the final PID parameters based on the basic PID parameters and the risk deviation coefficient include: Based on the trend level and the PID parameter range, determine the basic PID parameters; Based on the risk deviation coefficient, the risk urgency of the current basic PID parameters corresponding to the position within the buffer zone is quantified, and the control intensity of the basic PID parameters is dynamically amplified / reduced to determine the final PID parameters.

7. The compressor dynamic protection control method according to claim 6, characterized in that, In the step of quantifying the risk urgency of the current basic PID parameters corresponding to the position within the buffer zone based on the risk deviation coefficient, the formula for calculating the risk urgency is as follows: Risk deviation coefficient = (upper limit between buffer zones - actual value) / (upper limit between buffer zones - lower limit between buffer zones); Corrected deviation = Traditional deviation × (2 - Risk deviation coefficient).

8. An electronic starter, characterized in that, For implementing the compressor dynamic protection control method as described in any one of claims 1-7, the electronic starter includes: The segmentation module is used to tag the compressor's operating data into different scenarios, classify them into various load scenarios, and set basic critical value offset coefficients for each type of load scenario. The calculation module is used to collect compressor parameters in real time and calculate the real-time load intensity through a weighted algorithm. The parameters include at least the operating current, winding temperature and refrigerant pressure. The first determining module is used to determine the buffer zone based on the traditional critical value, the load intensity, and the basic critical value offset coefficient. The input module is used to acquire compressor data and convert it into a time-series feature sequence. The data in the time-series feature sequence is normalized based on the buffer space. Then, it is input into a pre-trained neural network model with a dual-branch network structure, which outputs the trend level and PID parameter range. The compressor data includes the operating current, the winding temperature, the refrigerant pressure, the outdoor ambient temperature, the indoor set temperature, the fan speed, the rate of change of the winding temperature within a preset time, and the rate of change of the operating current within a preset time. The neural network model with a dual-branch network structure includes a trend prediction branch and a PID parameter generation branch. The trend prediction branch is an LSTM layer + Attention network, and the PID parameter generation branch is a fully connected layer + residual network network. In addition, when data falls into the buffer space, the Attention mechanism assigns higher weights to the corresponding data. The second determining module is used to determine the basic PID parameters based on the trend level and the PID parameter range, introduce a risk deviation coefficient, and determine the final PID parameters based on the basic PID parameters and the risk deviation coefficient. The setting module is used to set a smooth transition limit for the output of the final PID parameters, and to run the compressor under the smooth transition limit.

Images