A method, device, storage medium, and electronic starter for intelligent control of a compressor.

Abstract

The application discloses a kind of compressor intelligent control method, device, storage medium and electronic control, it is related to compressor technical field, the method includes: the method includes: when compressor power on, the power-on parameter of corresponding compressor is obtained;Call built-in parameter matching model, and determine corresponding starting parameter and protection parameter by parameter matching model according to power-on parameter;Compressor is controlled using starting parameter and protection parameter.The application solves the problem of poor versatility of the existing technology in the starter, protector and compressor.

Description

Technical Field

[0001] This invention relates to the field of compressor technology, and in particular to a compressor intelligent control method, device, storage medium, and electronic starter. Background Technology

[0002] Traditional compressor starters and protectors are usually designed and parameter-matched for specific compressor models. This means that in practical applications, if a different compressor specification is required, the corresponding starter and protector must be replaced, which is a cumbersome and costly process. Summary of the Invention

[0003] In view of this, the purpose of the present invention is to provide a compressor intelligent control method, device, storage medium and electronic starter, which aims to solve the problem of poor versatility between starters, protectors and compressors in the prior art.

[0004] This invention proposes an intelligent control method for a compressor, the method comprising:

[0005] Once the compressor is powered on, obtain the corresponding power-on parameters of the compressor;

[0006] The built-in parameter matching model is invoked to determine the corresponding startup and protection parameters based on the power-on parameters.

[0007] The compressor is controlled using startup and protection parameters.

[0008] Furthermore, in the above-mentioned intelligent compressor control method, the step of obtaining the corresponding compressor power-on parameters after the compressor is powered on includes:

[0009] Upon receiving the compressor's power-on signal, it enters the pre-power-on mode and locks the basic parameter acquisition window period within a preset time period.

[0010] The mains power is stepped down to a preset voltage using a voltage regulation module, and then the required base power parameters are collected and temporarily stored in a temporary data buffer.

[0011] Switch to normal start mode, input rated voltage to compressor, and trigger transient acquisition window period of preset time period to collect transient characteristic parameters that characterize the fluctuation of basic ground parameters and temporarily store them in temporary data buffer area.

[0012] Switch to steady-state monitoring mode and trigger a preset stability verification window period to collect stability verification parameters that characterize the stability of the base power parameters and temporarily store them in a temporary data cache area.

[0013] Based on the power-on parameters, transient characteristic parameters, and stability verification parameters cached in the temporary data buffer, the corresponding power-on parameters of the compressor are obtained according to preset rules.

[0014] Furthermore, in the above-mentioned intelligent control method for compressors, the basic electrical parameters include average current, average voltage and ambient temperature, the transient characteristic parameters include peak current, rise time of current to peak current, voltage fluctuation and vibration frequency, and the stability verification parameters include current fluctuation range.

[0015] The step of obtaining the corresponding compressor power-on parameters according to preset rules based on the power-on parameters, transient characteristic parameters, and stability verification parameters cached in the temporary data cache includes:

[0016] The temperature-compensated resistance is obtained based on the average current, average voltage, and ambient temperature.

[0017] During the stability verification window, the peak values ​​of resistance and current after temperature compensation are obtained after a preset number of samplings, and the relative deviation of resistance and current after temperature compensation for each sampling is calculated.

[0018] When the relative deviation is less than the threshold, the resistance and current peak values ​​after temperature compensation are determined to be the final resistance and current peak values ​​after temperature compensation.

[0019] The resistance, peak current, rise time of current to peak value, vibration frequency, and current fluctuation range after temperature compensation are used as the power-on parameters.

[0020] Furthermore, in the above-mentioned intelligent compressor control method, the training process of the parameter matching model includes:

[0021] A training dataset is formed by acquiring a preset number of historical power-on parameters and their corresponding startup and protection parameters;

[0022] Construct a pre-defined neural network, input the training dataset into the pre-defined neural network for deep learning training until the loss function tends to stabilize to obtain the parameter matching model.

[0023] Furthermore, in the above-mentioned intelligent compressor control method, after the step of entering the pre-power-on mode and locking the basic parameter acquisition window period within a preset time period upon receiving the compressor's power-on signal, the method further includes:

[0024] Collect the real-time rate of change of the current and determine whether the real-time rate of change of the current is less than a threshold.

[0025] If so, the basic parameter acquisition window period will be ended early;

[0026] If not, then extend the basic parameter acquisition window period.

[0027] Furthermore, in the aforementioned intelligent compressor control method, after the step of calling the built-in parameter matching model and determining the corresponding start-up parameters and protection parameters based on the power-on parameters through the parameter matching model, the method further includes:

[0028] The changes in power-on parameters are obtained, and each change in power-on parameters is divided into a preset number of state intervals to obtain a preset number of discrete states.

[0029] Obtain the fine-tuning amount of the starting parameters and protection parameters corresponding to the change in power-on parameters, and obtain a preset number of action combinations based on the fine-tuning amount of the starting parameters and protection parameters;

[0030] Establish a corresponding reward table based on discrete states and action combinations, and initialize the reward value corresponding to each state and action combination;

[0031] The changes in power-on parameters during multiple compressor starts are collected, and these changes are mapped to preset discrete states to form the current environmental state.

[0032] Iterate through a preset number of action combinations, check the attempt status and reward value of each action combination, and add the corresponding action combination to the priority pool.

[0033] Generate random numbers within a preset range. Based on the random numbers and the exploration probability, determine whether to select the corresponding action combination from the priority pool or from all actions. Explore based on preset rules to obtain the optimal action combination.

[0034] The fine-tuning amount of the corresponding start-up parameters and protection parameters is determined based on the selected final action combination, and the start-up parameters and protection parameters are fine-tuned according to the fine-tuning amount.

[0035] Furthermore, in the aforementioned intelligent compressor control method, the step of generating random numbers within a preset range and determining whether to select the corresponding action combination from a priority pool or from all actions based on the random numbers and the exploration probability includes:

[0036] An initial random number is generated based on a random number generator. The operating condition bias coefficient is obtained to calculate the random number correction bias. The initial random number is then corrected using the random number correction bias to obtain the final random number.

[0037] The operating condition bias coefficients include aging bias coefficients, startup success rate bias coefficients, and exploration progress bias coefficients. Random number correction biases are determined based on the aging bias coefficients, startup success rate bias coefficients, and exploration progress bias coefficients.

[0038] Another object of the present invention is to provide a compressor intelligent control device, the device comprising:

[0039] The acquisition module is used to acquire the corresponding power-on parameters of the compressor after it is powered on.

[0040] The matching module is used to call the built-in parameter matching model and determine the corresponding start-up parameters and protection parameters based on the power-on parameters.

[0041] The control module is used to control the compressor using startup and protection parameters.

[0042] Another object of the present invention is to provide a readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described above.

[0043] Another object of the present invention is to provide an electronic starter, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the program to implement the steps of the method described above.

[0044] This invention obtains the compressor's power-on parameters after power-on; it then calls a built-in parameter matching model to determine the corresponding start-up and protection parameters based on the power-on parameters; and finally, it controls the compressor using these parameters. It can determine suitable start-up and protection parameters based on the identified compressor power-on parameters. This solves the problem of poor compatibility between starters, protectors, and compressors in existing technologies. Attached Figure Description

[0045] Figure 1 This is a flowchart of the intelligent compressor control method in the first embodiment of the present invention;

[0046] Figure 2 This is a structural block diagram of the intelligent control device for the compressor in the third embodiment of the present invention. Detailed Implementation

[0047] 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.

[0048] 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.

[0049] 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.

[0050] Example 1

[0051] Please see Figure 1 The figure shows a compressor intelligent control method in the first embodiment of the present invention, the method including steps S10 to S12.

[0052] Step S10: After the compressor is powered on, obtain the corresponding power-on parameters of the compressor.

[0053] In the initial stage of compressor power-on, high-precision sensors, timing control logic, and data preprocessing algorithms are used to systematically collect multi-dimensional key parameters that reflect compressor characteristics, providing accurate and reliable raw data support for subsequent parameter matching models.

[0054] Specifically, upon receiving the compressor's power-on signal, it enters the pre-power-on mode and locks in the basic parameter acquisition window for a preset time period;

[0055] The mains power is stepped down to a preset voltage using a voltage regulation module, and then the required base power parameters are collected and temporarily stored in a temporary data buffer.

[0056] Switch to normal start mode, input rated voltage to compressor, and trigger transient acquisition window period of preset time period to collect transient characteristic parameters that characterize the fluctuation of basic ground parameters and temporarily store them in temporary data buffer area.

[0057] Switch to steady-state monitoring mode and trigger a preset stability verification window period to collect stability verification parameters that characterize the stability of the base power parameters and temporarily store them in a temporary data cache area.

[0058] Based on the power-on parameters, transient characteristic parameters, and stability verification parameters cached in the temporary data buffer, the corresponding power-on parameters of the compressor are obtained according to preset rules.

[0059] During the pre-power-on mode (typically outputting 10% of the rated voltage, such as 22V under a 220V power grid), the pre-excitation current, winding terminal voltage and ambient temperature are mainly collected.

[0060] After the pre-power-on test shows no abnormalities, the system switches to normal start-up mode, inputs the rated voltage to the compressor, and triggers the start-up impact phase (usually lasting 50-300ms). During this phase, a high-frequency current sensor collects the peak value of the starting current and the time it takes for the current to rise to the peak value. At the same time, a vibration sensor collects the vibration frequency at the moment of start-up, capturing the essential characteristics of the compressor's mechanical structure (such as rotary or scroll compressors). After the start-up impact phase ends, the system enters the quasi-steady-state monitoring phase (usually lasting 300ms-2s). During steady-state operation, the system collects the current fluctuation range and real-time winding temperature, and performs temperature compensation on the winding resistance calculated in the pre-power-on phase to obtain the temperature compensation resistance at a standard 25℃, ultimately forming a complete set of power-on parameters.

[0061] For example, the basic grounding parameters include average current, average voltage and ambient temperature, transient characteristic parameters include peak current, rise time of current to peak current, voltage fluctuation and vibration frequency, and stability verification parameters include current fluctuation range.

[0062] The step of obtaining the corresponding compressor power-on parameters according to preset rules based on the power-on parameters, transient characteristic parameters, and stability verification parameters cached in the temporary data cache includes:

[0063] The temperature-compensated resistance is obtained based on the average current, average voltage, and ambient temperature.

[0064] During the stability verification window, the peak values ​​of resistance and current after temperature compensation are obtained after a preset number of samplings, and the relative deviation of resistance and current after temperature compensation for each sampling is calculated.

[0065] When the relative deviation is less than the threshold, the resistance and current peak values ​​after temperature compensation are determined to be the final resistance and current peak values ​​after temperature compensation.

[0066] The resistance, peak current, rise time of current to peak value, vibration frequency, and current fluctuation range after temperature compensation are used as the power-on parameters.

[0067] First, the "average current" in the basic power-on parameters is the average current value of continuous sampling in the pre-power-on mode (e.g., 10μs / sampling, the average of 5000 sampling points within a 50ms window), the "average voltage" is the average low voltage output of the voltage regulation module in this stage (e.g., the average value within the range of 22V±2V to verify voltage regulation accuracy), and the "ambient temperature" is the ambient temperature around the starter collected by the sensor during pre-power-on (to provide an initial temperature reference for subsequent resistance temperature compensation). The three together constitute the basic data for resistance calculation.

[0068] Among the transient characteristic parameters, "peak current" is the maximum instantaneous value of the current soaring from 0 to the peak value under normal startup mode (rated voltage input), "rise time of current to peak value" is the duration of current rising to peak value, "voltage fluctuation" is the maximum deviation of rated voltage during this stage (such as the difference between 220V mains voltage fluctuation and 200V, to judge the impact of grid stability on transient parameters), and "vibration frequency" is the characteristic vibration frequency of the compressor casing during startup impact. These parameters directly reflect the dynamic characteristics of the compressor during startup.

[0069] The "current fluctuation range" parameter in the stability verification parameters is the percentage of the difference between the maximum and minimum current values ​​from 3-5 consecutive samplings under steady-state monitoring mode relative to the average current, verifying operational stability. Next, the calculation logic for the "temperature-compensated resistance" is clarified—this resistance is the standardized resistance of the compressor winding at a standard 25°C (eliminating the influence of temperature on the resistance value and ensuring parameter comparability under different environments). The calculation first obtains the original winding resistance through the "average voltage" and "average current" during the pre-power-on phase, and then performs temperature compensation based on the "ambient temperature." This resistance is a core indicator reflecting the compressor's power specifications.

[0070] Next is the core deviation judgment rule for stability verification. To avoid interference with the peak resistance and current after temperature compensation in a single acquisition (such as power grid fluctuations or sensor noise), multiple samplings are required within the steady-state monitoring window (the preset number of samplings is usually 3, balancing efficiency and reliability). The relative deviation between each sampled value and the average of the three samples is calculated. For example, after three temperature compensations, the resistance sampled values ​​are 105.7Ω, 106.2Ω, and 105.9Ω, with an average of 105.9Ω. The maximum single relative deviation is (106.2-105.9) / 105.9×100%≈0.28%. If the preset threshold is 1%, then the deviation is less than the threshold, and the parameter is determined to be stable. If the relative deviation of a certain sampling exceeds the threshold, resampling is required until all three deviations meet the requirements, ensuring the reliability of the final parameter.

[0071] Once stability verification is passed, the "stable temperature-compensated resistance" and "stable peak current" (e.g., 4A collected during the transient phase, with a deviation of <1% after 3 verifications) are selected. These are then combined with the "rise time of current to peak value" (e.g., 80ms), "vibration frequency" (e.g., 50Hz) collected during the transient phase, and the "current fluctuation range" (e.g., 4.4%) during the steady-state phase. These five parameters together constitute the final "compressor power-on parameters." This parameter set includes standardized parameters reflecting the compressor's essential specifications (temperature-compensated resistance), transient parameters reflecting the startup dynamic characteristics (peak current, rise time, vibration frequency), and steady-state parameters reflecting operational stability (current fluctuation range). This provides a comprehensive and reliable basis for the subsequent parameter matching model to accurately output control parameters such as startup capacitor and overcurrent threshold, avoiding adaptation errors caused by missing or distorted parameters.

[0072] Step S11: Call the built-in parameter matching model and determine the corresponding startup parameters and protection parameters based on the power-on parameters.

[0073] First, the "built-in parameter matching model" is not a simple formula calculation, but a lightweight hybrid model built on "pre-training with massive industry compressor data + physical characteristic rule constraints" (usually a fully connected neural network combined with a rule engine, with the number of parameters controlled at the thousands level, adapting to the limited computing power and storage resources of the starter chip). The training data covers "power-on parameters - optimal start / protection parameters" sample pairs of all mainstream compressor types (rotary and scroll) from 100W to 5kW under different operating conditions (normal temperature, low temperature, power grid fluctuations). At the same time, it incorporates electrical physics rules (such as the positive correlation between starting capacitor and motor torque, and the proportional constraint between overcurrent threshold and current peak value) to avoid the model outputting parameters that violate physical common sense. The model is already fixed in the starter's Flash storage area before leaving the factory, and no on-site training is required. It only needs to be triggered by the real-time input power-on parameters to ensure response speed (single calculation time <10ms, without affecting the compressor start-up sequence).

[0074] Secondly, the power-on parameters received by the model need to be preprocessed to ensure that the input format is consistent with that during training. First, each parameter is standardized to eliminate the interference of dimensional differences on model calculation. Then, through "parameter validity verification", if a parameter is abnormal, a backup value (estimated based on other normal parameters) is used to avoid abnormal data causing deviation in model output.

[0075] For example, the training process of the parameter matching model includes:

[0076] A training dataset is formed by acquiring a preset number of historical power-on parameters and their corresponding startup and protection parameters;

[0077] Construct a pre-defined neural network, input the training dataset into the pre-defined neural network for deep learning training until the loss function tends to stabilize to obtain the parameter matching model.

[0078] The first step is to construct the training dataset, which is the foundation for model training. The preset amount of historical data usually needs to cover 10,000+ sets of samples (too few samples will lead to overfitting of the model and failure to generalize to different models of compressors). Each set of samples contains "input features (historical electrical parameters)" and "labels (corresponding start-up parameters and protection parameters)". The input features must be completely consistent with the dimensions of the compressor data with the actual collected electrical parameters. Finally, a uniformly formatted training dataset is formed and divided into a training set (for model learning), a validation set (for verifying accuracy during training), and a test set (for final evaluation of the model's generalization ability) in a 7:2:1 ratio.

[0079] Secondly, the construction of the pre-defined neural network needs to balance "model accuracy" and "lightweight requirements for embedded deployment." Instead of choosing complex deep networks (such as CNNs and LSTMs), a "fully connected neural network (FCN)" is constructed as the basic architecture. This typically includes a three-layer structure: input layer, hidden layer, and output layer. The number of neurons in the input layer matches the dimensionality of the power-on parameters. The hidden layers are designed with 2-3 layers (to avoid overfitting and wasted computational power due to too many layers), with 16-32 neurons per layer (experiments have verified that this number ensures good fitting ability while controlling the number of parameters to the thousands, adapting to the computing power of the starter chip). The live function is ReLU (to avoid gradient vanishing and for high computational efficiency); the number of neurons in the output layer is consistent with the dimension of the control parameters (4 neurons, corresponding to the start-up capacitance, start-up time, overcurrent threshold, and overheat threshold respectively). The activation function is selected according to the parameter type (linear activation is used for start-up capacitance / start-up time to retain the continuous numerical range; sigmoid activation is used for overcurrent threshold / overheat threshold to map to the actual physical range, such as mapping the overcurrent threshold to 1.5-4.5In). At the same time, a "Dropout layer" (dropout probability set to 0.1-0.2) is added to the network to prevent overfitting, and a "Batch Normalization layer" is added to accelerate training convergence and ensure that the model has sufficient mapping accuracy with a limited number of parameters.

[0080] Finally, there's the iterative process of deep learning training, the core of which is "optimizing model parameters through a loss function until the model stably outputs accurate labels": First, the pre-defined training set is input into a pre-defined neural network in batches. The network calculates the "predicted startup / protection parameters" through forward propagation. Next, the loss between the predicted values ​​and the true labels (the optimal parameters in the training set) is calculated. The loss function chosen is the "weighted mean square error (MSE)" (because different control parameters have different importance; for example, the startup capacitor has the greatest impact on the startup success rate, so its weight is set to 0.4; the overcurrent threshold is second, so its weight is set to 0.3; the startup time and overheat threshold are each weighted at 0.15 to avoid the error of any one parameter dominating the loss). Then, a backpropagation algorithm (such as the Adam optimizer, with the initial learning rate set to 0.001, decaying with each training epoch to avoid the error of a single parameter dominating the loss) is used. The model calculates the gradient of the weights of neurons in each layer (periodic oscillation) and updates the weights in the direction of gradient descent. After each epoch of training, the model performance is evaluated using a validation set (the validation set loss and parameter prediction error are calculated). If the validation set loss does not decrease for 5-10 consecutive epochs (i.e., it tends to stabilize, usually with a loss fluctuation of <0.001), training is stopped. At this point, the model has fully learned the mapping relationship between the power-on parameters and the control parameters. Finally, the model's generalization ability is evaluated using a test set. If the parameter prediction error on the test set is <3% (e.g., the deviation between the predicted value and the true value of the start-up capacitor is <1μF), the model is considered to have passed training and can be embedded into the starter chip. If the error exceeds the standard, more diverse training data needs to be added, the network structure needs to be adjusted (e.g., the number of neurons in the hidden layers needs to be increased or decreased), and the model needs to be retrained until the accuracy requirements are met. The entire training process can be completed on a server with sufficient computing power (rather than on the initiator itself). After training, the model is quantized (e.g., quantizing 32-bit floating-point weights into 16-bit weights to reduce storage usage) to ensure that the model can adapt to the hardware resources of the initiator's embedded chip while maintaining high-precision parameter mapping capabilities.

[0081] Step S12: Control the compressor using startup parameters and protection parameters.

[0082] In summary, the intelligent compressor control method in the above embodiments of the present invention obtains the corresponding compressor power-on parameters after the compressor is powered on; calls the built-in parameter matching model, and determines the corresponding start-up parameters and protection parameters based on the power-on parameters; and controls the compressor using the start-up parameters and protection parameters. It can determine suitable start-up parameters and protection parameters based on the identified compressor power-on parameters. This solves the problem of poor versatility between starters, protectors, and compressors in the prior art.

[0083] Example 2

[0084] This embodiment also proposes a compressor intelligent control method. The difference between the compressor intelligent control method in this embodiment and the compressor intelligent control method in Embodiment 1 is as follows:

[0085] The step of calling the built-in parameter matching model and determining the corresponding startup and protection parameters based on the power-on parameters through the parameter matching model also includes:

[0086] The changes in power-on parameters are obtained, and each change in power-on parameters is divided into a preset number of state intervals to obtain a preset number of discrete states.

[0087] Obtain the fine-tuning amount of the starting parameters and protection parameters corresponding to the change in power-on parameters, and obtain a preset number of action combinations based on the fine-tuning amount of the starting parameters and protection parameters;

[0088] Establish a corresponding reward table based on discrete states and action combinations, and initialize the reward value corresponding to each state and action combination;

[0089] The changes in power-on parameters during multiple compressor starts are collected, and these changes are mapped to preset discrete states to form the current environmental state.

[0090] Iterate through a preset number of action combinations, check the attempt status and reward value of each action combination, and add the corresponding action combination to the priority pool.

[0091] Generate random numbers within a preset range. Based on the random numbers and the exploration probability, determine whether to select the corresponding action combination from the priority pool or from all actions. Explore based on preset rules to obtain the optimal action combination.

[0092] The fine-tuning amount of the corresponding start-up parameters and protection parameters is determined based on the selected final action combination, and the start-up parameters and protection parameters are fine-tuned according to the fine-tuning amount.

[0093] Based on the parameter matching model determining the initial start-up / protection parameters, the start-up and protection parameters are further dynamically fine-tuned through lightweight reinforcement learning logic. The core solution is to address the initial parameter mismatch problem caused by aging and operating condition fluctuations (such as changes in grid voltage and sudden changes in ambient temperature) throughout the compressor's life cycle. First, the premise of the entire process is that "the parameter matching model has output the initial start-up parameters (such as a start-up capacitor of 35μF and a start-up time of 100ms) and protection parameters (such as an overcurrent threshold of 3.2In and an overheat threshold of 115℃)". Subsequent steps all revolve around "fine-tuning these initial parameters according to the changes in power-on parameters (such as the resistance increasing from 4.8Ω to 5.5Ω due to aging). In essence, it is a closed-loop application of "agent (starter) - environment (compressor operating condition) - action (parameter fine-tuning) - reward (adaptation effect)" in reinforcement learning. Moreover, all designs are adapted to the low computing power and small storage requirements of embedded chips.

[0094] To reduce computational complexity, the continuous "power-on parameter changes" need to be divided into a preset number of discrete state intervals (usually each parameter is divided into 10 intervals, forming 10×10×10=1000 discrete states). This can convert continuous data into integer states that are easy for embedded chips to process. At the same time, the "fine-tuning of startup / protection parameters" need to be combined into a preset number of action combinations.

[0095] Establish a reward table (i.e., Q-table in reinforcement learning) corresponding to the "discrete state-action combination". The table dimension is "number of discrete states × number of action combinations" (e.g., 1000 × 25). Each cell stores the initial reward value of the "state-action" (usually set to 0.5, with no initial bias). The reward value will be updated later according to the fine-tuning effect (e.g., whether the start is successful or whether the parameter fluctuation is reduced). The table is stored in the starter Flash.

[0096] Before each compressor start-up, the changes in "current power-on parameters and initial parameters" are collected and these continuous changes are mapped to a preset discrete state range to form the current environmental state, providing a basis for subsequent action selection.

[0097] To balance the "exploration (trying new actions)" and "exploitation (selecting known high-reward actions)" of reinforcement learning, high-value actions need to be selected and added to the priority pool: traverse 25 action combinations, check the "trial status" (0 = not tried, 1 = tried) and "current reward value" of each action, and add "untried actions" (to avoid missing potential effective actions) or "tried actions with reward value > 0.8" (to utilize known effective actions) to the priority pool. For example, out of 25 actions, there are 5 untried actions and 3 high-reward actions, a total of 8 actions entering the priority pool, reducing the probability of blind exploration.

[0098] The system pre-defines a numerical range (e.g., [0,1]) and generates a random number based on this range each time an action combination needs to be selected. Simultaneously, the system pre-defines an "exploration probability" (e.g., 0.2, representing a 20% exploration probability and an 80% utilization probability; this probability can be dynamically adjusted according to actual needs or learning progress). This probability is used to differentiate the triggering conditions for the two selection modes: "utilizing existing experience" and "exploring new actions." Next, the generated random number is compared with the preset exploration probability: if the random number is less than or equal to the exploration probability, the "exploration mode" is triggered. In this mode, the system randomly selects an action combination from all possible action combinations (regardless of whether these combinations have been verified or deemed optimal), or selects one according to specific rules. This involves attempting action combinations that haven't been fully explored, avoiding getting stuck in local optima due to limited experience. If the random number is greater than the exploration probability, the "exploitation mode" is triggered. In this mode, the system selects action combinations from a pre-built "priority pool." This "priority pool" is formed by filtering and sorting action combinations based on past experience (such as the execution effect, reward value, success rate, etc.). The action combinations in the pool are all high-quality action combinations that have been verified to have high priority (such as higher reward or higher success rate after execution). Selecting such action combinations ensures the effectiveness and stability of the current decision. The entire process achieves a balance between "utilizing existing high-quality actions to ensure current performance" and "exploring unknown actions to expand the optimization space" through dynamic judgment of random numbers and exploration probabilities, thus avoiding local optima.

[0099] Finally, based on the selected optimal action combination, the corresponding fine-tuning amount is extracted, superimposed on the initial startup / protection parameters, and fine-tuned through the starter hardware module. At the same time, the attempt status and startup effect (success / failure) of this action are recorded for subsequent update of the reward table, forming a learning loop of "status-action-reward".

[0100] Furthermore, the step of generating random numbers within a preset range, and determining whether to select the corresponding action combination from the priority pool or from all actions based on the random numbers and the exploration probability, includes:

[0101] An initial random number is generated based on a random number generator. The operating condition bias coefficient is obtained to calculate the random number correction bias. The initial random number is then corrected using the random number correction bias to obtain the final random number.

[0102] The operating condition bias coefficients include aging bias coefficients, startup success rate bias coefficients, and exploration progress bias coefficients. Random number correction biases are determined based on the aging bias coefficients, startup success rate bias coefficients, and exploration progress bias coefficients.

[0103] The initial random number (range [0,1)) is generated using a hardware random number generator built into the starter (such as the HRNG of STM32). Then, the "operating condition bias coefficient" is calculated: aging bias coefficient, start-up success rate bias coefficient (calculated based on the success rate of the last 10 starts; the lower the success rate, the more negative the coefficient; for example, a success rate of 30% results in a coefficient of -0.25, making the random number smaller and favoring the exploration of new actions), and exploration progress bias coefficient (calculated based on the percentage of unattended actions; the higher the percentage, the more negative the coefficient). If the coefficient is -0.2 when no attempt is made (80% of the time, accelerate exploration); then calculate the total random number correction bias using the three bias coefficients according to their weights (aging 0.4, success rate 0.3, exploration progress 0.3) (e.g., 0.4×0.3+0.3×(-0.25)+0.3×(-0.2)=0.045); finally, adjust the initial random number using the correction bias (e.g., initial random number=0.42, corrected=0.42×(1+0.045)=0.4389) to obtain the final random number. Then compare the final random number with the exploration probability (e.g., 0.2): if the random number < exploration probability (e.g., 0.4389<0.2 is not true), then select the action with the highest reward value from all actions (exploitation); if true, then select an action from the priority pool (exploration), and finally determine the best action combination.

[0104] In summary, the intelligent compressor control method in the above embodiments of the present invention obtains the corresponding compressor power-on parameters after the compressor is powered on; calls the built-in parameter matching model, and determines the corresponding start-up parameters and protection parameters based on the power-on parameters; and controls the compressor using the start-up parameters and protection parameters. It can determine suitable start-up parameters and protection parameters based on the identified compressor power-on parameters. This solves the problem of poor versatility between starters, protectors, and compressors in the prior art.

[0105] Example 3

[0106] Please see Figure 2 The image shows a compressor intelligent control device proposed in the third embodiment of the present invention. The device includes:

[0107] The acquisition module 100 is used to acquire the corresponding power-on parameters of the compressor after it is powered on.

[0108] The matching module 200 is used to call the built-in parameter matching model and determine the corresponding start-up parameters and protection parameters based on the power-on parameters.

[0109] The control module 300 is used to control the compressor using startup parameters and protection parameters.

[0110] The functions or operation steps implemented by the above modules are largely the same as those in the above method embodiments, and will not be repeated here.

[0111] Example 4

[0112] In another aspect, the present invention provides a readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the steps of the method described in any one of Embodiments 1 to 2 above.

[0113] Example 5

[0114] In another aspect, the present invention provides an electronic starter, which includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the program, it implements the steps of the method described in any one of the above embodiments one to two.

[0115] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0116] Those skilled in the art will understand that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequential list of executable instructions for implementing logical functions, and can be embodied in any computer-readable storage medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable storage medium" can mean any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0117] More specific examples (a non-exhaustive list) of computer-readable storage media include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable storage media can even be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0118] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0119] 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.

[0120] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, 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 modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method of intelligent control of a compressor, characterized in that, The method comprises: When the compressor is powered on, the corresponding power-on parameters of the compressor are acquired; An internal parameter matching model is called, and the corresponding starting parameters and protection parameters are determined through the parameter matching model according to the power-on parameters; The starting parameters and the protection parameters are used to control the compressor; The step of acquiring the corresponding power-on parameters of the compressor when the compressor is powered on comprises: After receiving the power-on signal of the compressor, a pre-power-on mode is entered, and a basic parameter acquisition window period of a preset time period is locked; The mains voltage is reduced to a preset voltage by using a voltage regulating module, and then the required basic power-on parameters are acquired and temporarily stored in a temporary data buffer area; Switch to the normal starting mode, input the rated voltage to the compressor, and trigger the transient acquisition window period of the preset time period, respectively acquire the transient characteristic parameters representing the fluctuation of the basic power-on parameters, and temporarily store them in the temporary data buffer area; Switch to the steady-state monitoring mode, and trigger the stability verification window period of the preset time period, respectively acquire the stability verification parameters representing the stability of the basic power-on parameters, and temporarily store them in the temporary data buffer area; According to the basic power-on parameters, the transient characteristic parameters and the stability verification parameters cached in the temporary data buffer area, the corresponding power-on parameters of the compressor are obtained according to the preset rules.

2. The intelligent control method of a compressor according to claim 1, wherein The basic power-on parameters include average current, average voltage and environmental temperature, the transient characteristic parameters include current peak value, current rising time to peak value, voltage fluctuation and vibration frequency, and the stability verification parameters include current fluctuation range; The step of obtaining the corresponding power-on parameters of the compressor according to the basic power-on parameters, the transient characteristic parameters and the stability verification parameters cached in the temporary data buffer area according to the preset rules comprises: According to the average current, the average voltage and the environmental temperature, a temperature-compensated resistance is obtained; In the stability verification window period, the temperature-compensated resistance and the current peak value of a preset sampling number are acquired, and the relative deviation of the temperature-compensated resistance and the current of each sampling is calculated; When the relative deviation is less than a threshold value, the temperature-compensated resistance and the current peak value are determined as the final temperature-compensated resistance and current peak value; The temperature-compensated resistance, the current peak value, the current rising time to peak value, the vibration frequency and the current fluctuation range are taken as the power-on parameters.

3. The intelligent control method of a compressor according to claim 2, wherein, The training process of the parameter matching model comprises: A preset number of historical power-on parameters and corresponding starting parameters and protection parameters are acquired to form a training data set; A preset neural network is constructed, and the training data set is input into the preset neural network for deep learning training until the loss function tends to be stable to obtain the parameter matching model.

4. The intelligent control method of a compressor according to claim 3, wherein After the step of entering the pre-power-on mode and locking the basic parameter acquisition window period of the preset time period when the power-on signal of the compressor is received, the following steps are further included: The real-time change rate of the current is acquired, and it is judged whether the real-time change rate of the current is less than a threshold value; If yes, the basic parameter acquisition window period is ended in advance; If not, the basic parameter acquisition window period is extended.

5. The intelligent control method of a compressor according to claim 4, wherein After the step of calling the internal parameter matching model, the corresponding starting parameters and protection parameters are determined through the parameter matching model according to the power-on parameters, the following steps are further included: The power-on parameter change amount is obtained, and the power-on parameter change amount is divided into a preset number of state intervals to obtain a preset number of discrete states; The adjustment amount of the starting parameter and the protection parameter corresponding to the power-on parameter change amount is obtained, and the adjustment amount of the starting parameter and the protection parameter is used to obtain a preset number of action combinations; A reward table corresponding to the discrete state and the action combination is established, and the reward value corresponding to each state and action combination is initialized; The power-on parameter change amount of multiple compressor starts is collected, and the power-on parameter change amount is mapped to the preset discrete state to form a current environment state; The preset number of action combinations is traversed, and the trial state and the reward value of each action combination are checked to add the corresponding action combination to a priority pool; A random number of a preset interval is generated, and it is determined whether to select the corresponding action combination from the priority pool or select the corresponding action combination from all actions based on the random number and an exploration probability, and the best action combination is obtained based on a preset rule; The adjustment amount of the starting parameter and the protection parameter corresponding to the selected final action combination is determined, and the starting parameter and the protection parameter are adjusted based on the adjustment amount.

6. The intelligent control method of a compressor according to claim 5, wherein, The step of generating a random number of a preset interval and determining whether to select the corresponding action combination from the priority pool or select the corresponding action combination from all actions based on the random number and an exploration probability includes: An initial random number is generated based on a random number generator, a working condition bias coefficient is obtained to calculate a random number correction bias, and the initial random number is corrected based on the random number correction bias to obtain a final random number; The working condition bias coefficient includes an aging bias coefficient, a starting success rate bias coefficient, and an exploration progress bias coefficient, and the random number correction bias is determined based on the aging bias coefficient, the starting success rate bias coefficient, and the exploration progress bias coefficient.

7. A compressor intelligent control device, characterized by, The device for implementing the compressor intelligent control method of any one of claims 1 to 6 includes: An acquisition module for acquiring the power-on parameters of the corresponding compressor after the compressor is powered on; A matching module for calling a built-in parameter matching model and determining the starting parameter and the protection parameter based on the power-on parameters through the parameter matching model; A control module for controlling the compressor based on the starting parameter and the protection parameter.

8. A readable storage medium, having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the steps of the method of any one of claims 1 to 6.

9. An electronic actuator, characterized in that The device includes a memory, a processor, and a computer program stored in the memory and running on the processor, and the processor implements the steps of the method of any one of claims 1 to 6 when executing the program. The device includes a memory, a processor, and a computer program stored in the memory and running on the processor, and the processor implements the steps of the method of any one of claims 1 to 6 when executing the program.

Images