Control method of a heat pump system and computing device

Abstract

The application discloses a control method of a heat pump system, a computing device and a readable storage medium, comprising: collecting an operation signal of a vehicle; performing data preprocessing on the operation signal to obtain time sequence input data; based on a preset time sequence prediction neural network model, outputting a time sequence prediction result of the heat pump system in a future preset time domain according to the time sequence input data; performing MPC multi-objective rolling optimization solving under a preset constraint condition according to the time sequence prediction result to obtain a target control sequence; and controlling a corresponding actuator to execute a corresponding cooperative action instruction according to the target control sequence. In this way, the traditional lag feedback control is changed into feedforward active adjustment, cabin temperature overshoot and oscillation are significantly inhibited, and temperature control response speed and stability accuracy are improved.

Description

Technical Field

[0001] This application relates to the field of intelligent control of automotive thermal management systems, and specifically to a control method for a heat pump system. Background Technology

[0002] With the continuous improvement of the intelligence level of new energy vehicles, all actuators in the vehicle complete precise responses under the unified scheduling of the electronic control unit. Electrification significantly improves the control accuracy and response speed of components, better meeting the needs of users in various scenarios, but it also poses greater challenges to the complexity and adaptability of control logic. Currently, the control algorithm of the vehicle thermal management system mainly relies on traditional PID control or physical model-based model predictive control (MPC), which completes the coordinated control of actuators under different operating conditions through preset software models.

[0003] However, as the number of actuators in the thermal management system continues to increase, the control requirements for multi-mode switching and multi-dynamic scenarios become increasingly complex. Existing control schemes generally have the following problems: control oscillations lead to insufficient temperature control accuracy, the mode switching process generates significant impact, the target response is delayed, the overall system energy consumption is too high, and it is difficult to effectively cope with dynamic working conditions with strong nonlinearity and multi-physics coupling.

[0004] Therefore, how to provide a control method for the thermal management system of new energy vehicles that can integrate timing prediction and rolling optimization, achieve multi-actuator collaborative control, and has strong robustness has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0005] The purpose of this application is to provide a control method, computing device and readable storage medium for a heat pump system, which can transform traditional hysteresis feedback control into feedforward active regulation, significantly suppress cabin temperature overshoot and oscillation, and improve temperature control response speed and stability accuracy.

[0006] To achieve the above objectives: In a first aspect, embodiments of this application provide a control method for a heat pump system, the method comprising: Collect vehicle operating signals; The operating signals are preprocessed to obtain timing input data; Based on a preset time-series prediction neural network model, the system outputs time-series prediction results for the heat pump system within a preset time domain, according to the time-series input data. Based on the time series prediction results, MPC multi-objective rolling optimization is performed under preset constraints to obtain the target control sequence. Based on the target control sequence, control the corresponding actuator to execute the corresponding coordinated action instructions.

[0007] In one embodiment, the running signal is preprocessed to obtain timing input data, including: The raw data of the operating signal is subjected to time alignment, filtering and noise reduction, missing value interpolation and normalization to obtain the standard signal of the heat pump system. The standard signal and its corresponding timestamp are framed by sliding within a preset window length to obtain the timing input data; the timing input data includes the current time and historical data from multiple past times.

[0008] In one embodiment, based on a preset time-series prediction neural network model, and according to the time-series input data, the system outputs a time-series prediction result for the future preset time domain of the heat pump system, including: Input the time-series input data into a preset time-series prediction neural network model to predict the state change trend of the heat pump system within a preset time domain in the future; Based on the trend of state changes, the system outputs the time-series prediction results for the future preset time domain of the heat pump system.

[0009] In one embodiment, before obtaining the target control sequence by performing MPC multi-objective rolling optimization solution under preset constraints based on the time series prediction results, the method further includes: Determine the current operating mode of the heat pump system based on standard signals; Set the preset initial values ​​for MPC multi-objective rolling solution based on the target working condition mode.

[0010] In one embodiment, based on the time-series prediction results, MPC multi-objective rolling optimization is performed under preset constraints to obtain the target control sequence, including: Based on the preset initial values ​​and timing prediction results, the multi-objective solution is performed using the MPC multi-objective rolling optimization algorithm under preset constraints, and the target control sequence of each actuator in the control time domain is obtained.

[0011] In one embodiment, outputting the target control sequence to the corresponding actuator to execute the corresponding cooperative action instruction includes: Based on the preset data arrangement rules of the target control sequence, determine the position index number of each value in the control quantity sequence at the current moment; Based on the correspondence between position index number and actuator type, the target actuator category and its control parameters matched for each control variable are obtained; Based on the target actuator type, the corresponding actuator driver interface is called to convert the control quantity into a drive signal and output it, thus obtaining the coordinated action command of multiple actuators in the heat pump system.

[0012] In one embodiment, the method further includes: Based on the actual system response data collected after the control input is output to the corresponding actuator, the real-time measured value of the current status is obtained. Based on the real-time measured values ​​and the preset "first-in-first-out" update rule of the sliding window, the update result of the sliding window is obtained; Based on the update results, the historical data sequence used in the next control time domain is obtained.

[0013] In one embodiment, the method further includes: performing a rationality diagnosis on the prediction results of the time-series prediction neural network; if the diagnosis result is abnormal, generating a first abnormality monitoring signal. The convergence status of the MPC multi-objective optimization solution is monitored. If non-convergence is detected, a second abnormal monitoring signal is generated. When any of the above monitoring signals is triggered, the control system automatically switches from the current timing prediction-MPC cooperative control mode to the preset PID fallback control mode.

[0014] Secondly, embodiments of this application provide a computing device, including: a processor and a memory storing a computer program, wherein when the processor runs the computer program, the steps of the control method for the heat pump system described above are implemented.

[0015] Thirdly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the control method for the heat pump system described above.

[0016] The control method, computing device, and readable storage medium for a heat pump system provided in this application embodiment include: firstly, acquiring vehicle operating signals and preprocessing the signals to obtain time-series input data; then, based on a preset time-series prediction neural network model, outputting a time-series prediction result for the heat pump system within a preset time domain based on the time-series input data; based on the time-series prediction result, performing MPC multi-objective rolling optimization under preset constraints to obtain a target control sequence; and finally, controlling the corresponding actuators to execute corresponding coordinated action commands based on the target control sequence. Thus, by acquiring vehicle operating signals and constructing time-series inputs, and using a time-series prediction neural network to accurately predict future heat load changes, traditional hysteresis feedback control is transformed into feedforward active adjustment, significantly suppressing cabin temperature overshoot and oscillation, and improving temperature control response speed and stability accuracy; based on the MPC rolling optimization framework, multi-objective global optimal solutions are solved online under multiple constraints, achieving a dynamic balance between heating performance, energy consumption, and comfort, effectively improving system energy efficiency and reducing low-temperature range degradation rate. Meanwhile, each control cycle is continuously revised based on the latest predictive information, endowing the system with strong anti-disturbance capabilities and adaptively responding to sudden changes in vehicle speed and environment; multi-actuator decoupling and synchronization eliminate action conflicts, ensuring reliability and durability; and by predicting extreme boundaries such as exhaust temperature and overheating, proactive safety avoidance measures are taken to enhance redundancy in extreme operating conditions. Ultimately, this achieves synergistic optimization of comfort, economy, and reliability throughout the entire life cycle, demonstrating significant engineering application value. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the control method for a heat pump system provided in an embodiment of this application.

[0018] Figure 2 This is a system hardware architecture diagram of the control method for a heat pump system provided in an embodiment of this application.

[0019] Heat exchanger expansion valve BEXV, evaporator expansion valve EEXV, heat pump gas-side electronic expansion valve AEXV, battery water pump BCPF, heater water pump HCWP, motor loop water pump EDCP, front-end radiator / Fan, heater core, heater heat exchanger EVAP, heat exchanger chiller, pressure and temperature sensors PT1 / PT2, water temperature sensors WT1 / WT2 / WT3 / WT4 / WT5, heat exchanger water-side LCC, battery pack.

[0020] Figure 3 A schematic diagram of the structure of a computing device provided in an embodiment of this application.

[0021] Processor 310, memory 311, network interface 312, bus system 313. Detailed Implementation

[0022] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. In the following description relating to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements.

[0023] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, components, features, and elements with the same names in different embodiments of this application may have the same meaning or different meanings, the specific meaning of which must be determined by its interpretation in that specific embodiment or further in conjunction with the context of that specific embodiment.

[0024] It should be understood that although the terms first, second, third, etc., may be used herein to describe various information, such information should not be limited to these terms. These terms are used only to distinguish information of the same type from one another. For example, without departing from the scope of this document, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if," as used herein, can be interpreted as "when," "when," or "in response to determination." Furthermore, as used herein, the singular forms "a," "an," and "the" are intended to also include the plural forms unless the context indicates otherwise. It should be further understood that the terms "comprising," "including," indicate the presence of the stated feature, step, operation, element, component, item, kind, and / or group, but do not exclude the presence, occurrence, or addition of one or more other features, steps, operations, elements, components, items, kinds, and / or groups. The terms "or" and "and / or" as used herein are to be interpreted as inclusive, or mean any one or any combination thereof. Therefore, "A, B, or C" or "A, B, and / or C" means "any one of the following: A; B; C; A and B; A and C; B and C; A, B, and C". Exceptions to this definition will only occur if the combination of elements, functions, steps, or operations is inherently mutually exclusive in some way.

[0025] It should be understood that although the steps in the flowcharts of this application's embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some of the steps in the figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least a portion of the sub-steps or stages of other steps.

[0026] It should be noted that step designations such as S101 and S102 are used in this document for the purpose of more clearly and concisely describing the corresponding content, and do not constitute a substantial limitation on the order. In specific implementation, those skilled in the art may execute S102 first and then S101, etc., but these should all be within the protection scope of this application.

[0027] It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0028] In the following description, the use of suffixes such as "module," "part," or "unit" to denote elements is solely for the purpose of illustrative purposes and has no specific meaning in itself. Therefore, "module," "part," or "unit" may be used interchangeably.

[0029] To facilitate understanding of this application, the following explanations are provided for the terms and technical objects that may be used in this application: Artificial Neural Networks (ANNs) are computational models inspired by the information processing mechanisms of biological neurons, consisting of a large number of interconnected artificial neurons. In this model, each neuron receives external input signals (such as physical quantities like temperature, pressure, and flow rate), performs weighted summation of the inputs, and then maps and transforms them through nonlinear activation functions (such as ReLU and Sigmoid) to ultimately generate the output result.

[0030] See Figure 1 This application provides a control method for a heat pump system. This method can be executed by a computing device provided in this application embodiment. The device can be implemented using software and / or hardware. In this embodiment, the computing device is taken as the executing entity of the method. The method provided in this embodiment includes: Step S101: Collect vehicle operation signals.

[0031] Among them, the operating signals can be a set of physical parameters used in the real-time interaction and decision-making between the vehicle controller, battery management system, motor controller, and thermal management system in the electronic control system of new energy vehicles (or intelligent connected vehicles). For example: vehicle signals (such as ambient temperature, vehicle speed, charging status, vehicle operating status, etc.), battery BMS signals (such as battery SOC, charging and discharging current, average battery temperature, maximum battery temperature, minimum battery temperature, etc.), motor signals (such as motor body temperature, motor current, motor voltage, etc.), cabin air conditioning signals (such as evaporator outlet air temperature, heater outlet air temperature, air conditioning temperature setting, air conditioning airflow setting, etc.), sensor signals (such as water temperature, compressor inlet and outlet pressure and temperature, etc.), actuator feedback signals (such as the actual feedback opening / speed of each actuator of the compressor, three expansion valves, three water pumps, water valve, and fan, and the actual current / voltage of the actuator, etc.).

[0032] Step S102: Perform data preprocessing on the running signals to obtain timing input data.

[0033] Data preprocessing includes time-series alignment, filtering and denoising (e.g., outlier detection, jump removal), missing value imputation, feature normalization, and time-series framing.

[0034] Among them, time-series input data refers to multivariate feature sequences that are strictly ordered in time and have a fixed time window length (such as 100ms) and are input into the time-series prediction neural network.

[0035] Step S103: Based on the preset time-series prediction neural network model, output the time-series prediction results of the heat pump system in the future preset time domain according to the time-series input data.

[0036] Temporal prediction neural network models refer to a class of artificial neural networks specifically designed for processing and analyzing data arranged in chronological order (time series). Their fundamental task is to accurately predict values ​​for one or more future time points by learning the time dependencies, dynamic evolution patterns, and periodic fluctuations inherent in historical data. For example, a temporal prediction neural network can be an LSTM (Long Short-Term Memory) neural network, a typical variant of a recurrent neural network (RNN) specifically designed for time series modeling and prediction. By introducing a gating mechanism, this model not only effectively avoids the gradient vanishing and gradient exploding problems common in traditional RNNs during long-sequence training but also significantly enhances the network's ability to capture long-term dependencies in time series data.

[0037] The preset time domain refers to the total length of time that the time-series prediction neural network model needs to cover in the future time axis when it outputs prediction results.

[0038] Among them, the time series prediction result refers to a series of predicted values ​​of how each output feature changes over time within a future preset time domain after the time series input data is input into the trained time series prediction neural network model.

[0039] Step S104: Based on the time series prediction results, perform MPC multi-objective rolling optimization under preset constraints to obtain the target control sequence.

[0040] Among them, the preset constraints refer to the limitations that the system must strictly adhere to during the MPC multi-objective rolling optimization solution process, which are determined by the physical hardware characteristics, safety protection mechanisms, and the inherent performance of the actuators. For example: expansion valve: opening degree 0%—100%, maximum adjustment step size for a single opening degree; Water pumps: their own upper and lower speed limits and speed regulation rate limits; Nine-way valve and four-way valve: mutually exclusive positions and time-delay constraint for reversing action; System safety constraints: refrigerant high and low pressure extremes, lower limit of superheat, safe range of battery temperature, upper and lower limits of water temperature in different circuits.

[0041] Among them, MPC multi-objective rolling optimization is a control strategy that, within each control cycle, uses the future time-series prediction results output by the time-series prediction neural network model as feedforward, and integrates the current real-time feedback state of the system. The MPC multi-objective solver, under the premise of satisfying the system's hard physical constraints (such as speed limits, pressure limits, temperature protection thresholds, etc.), collaboratively weighs multiple potentially conflicting control objectives to solve for a set of optimal control sequences within the future finite time domain. This strategy only executes the first step of the control command at the current moment, and recalculates the optimization in the next cycle based on the new system feedback, thereby achieving continuous rolling updates and closed-loop correction.

[0042] In this context, the target control sequence refers to the optimal set of actuator instructions arranged in chronological order, calculated by the optimization algorithm within a preset future control period during the multi-objective rolling optimization process (MPC). This set is based on the system's current real-time state, the future predicted trajectory provided by the time-series predictive neural network, and various physical constraints. Each element in this sequence corresponds to the setpoint or action of each actuator at a future discrete time point. However, in actual operation, only the first instruction in the sequence is issued and executed; subsequent instructions are recalculated in the next cycle as the system state is updated.

[0043] Step S105: According to the target control sequence, control the corresponding actuator to execute the corresponding cooperative action instruction.

[0044] The actuators may include the following: electric compressor, BEXV (Chiller heat exchanger expansion valve), EEXV (EVAP evaporator expansion valve), AEXV (heat pump gas-side electronic expansion valve), battery water pump BCPF, heater water pump HCWP, motor loop water pump EDCP, nine-way reversing valve mode opening, four-way reversing proportional valve opening, fan (fan) operating speed, etc.

[0045] The coordinated action instruction refers to the set of real-time target values ​​(such as the target speed of the compressor, the speed of each water pump, the opening degree of each valve, and the fan duty cycle) that are extracted from the complete target control sequence after the MPC multi-objective rolling optimization solver completes the optimization calculation for the current control cycle. This set is synchronously sent to each actuator. This instruction is the only actual output of the target control sequence in the current cycle. It carries the immediate result of the optimization decision and is also the starting signal that drives the execution layer response and realizes closed-loop feedback.

[0046] In summary, the method provided in the above embodiments involves: collecting vehicle operating signals; preprocessing the operating signals to obtain time-series input data; based on a preset time-series prediction neural network model, outputting the time-series prediction results of the heat pump system within a preset time domain based on the time-series input data; performing MPC multi-objective rolling optimization under preset constraints based on the time-series prediction results to obtain the target control sequence; and controlling the corresponding actuators to execute corresponding cooperative action commands based on the target control sequence. This transforms traditional hysteresis feedback control into feedforward active regulation, significantly suppressing cabin temperature overshoot and oscillation, and improving temperature control response speed and stability accuracy.

[0047] In one embodiment, the running signal is preprocessed to obtain timing input data, including: The raw data of the operating signal is subjected to time alignment, filtering and noise reduction, missing value interpolation and normalization to obtain the standard signal of the heat pump system. The standard signal and its corresponding timestamp are framed by sliding within a preset window length to obtain the timing input data; the timing input data includes the current time and historical data from multiple past times.

[0048] Timing alignment refers to the process of resampling and interpolating based on a preset unified control period (e.g., 100ms) to ensure that each row of data corresponds to the same time, thereby eliminating timing misalignment and constructing a fixed N-column input feature matrix, since the signal sampling periods of various controllers in the vehicle (such as BMS, MCU, VCU, ACU) are different (e.g., some signals have a period of 10ms, and some have a period of 1s).

[0049] Outlier detection refers to determining the validity of raw data based on the physical limits of each sensor's hardware. Values ​​exceeding the reasonable measurement range are discarded or marked. Specific threshold settings can be as follows: temperature: -40℃ to 130℃, valve opening: 0-100%, pressure: 0-35 bar, compressor speed: 0-8000 rpm, fan duty cycle: 0-100%, etc.

[0050] Among these, jump rejection refers to using a differential threshold method to identify interference jump points when the magnitude of data mutation exceeds the reasonable physical change rate of the signal within a short period of time, and then rejecting or marking them. Simultaneously, data lines with all zeros generated when the vehicle is powered off, in sleep mode, or when message transmission is interrupted are directly deleted to avoid interfering with subsequent analysis.

[0051] Among them, missing value interpolation processing refers to the hierarchical processing strategy for missing values ​​(NaN) caused by scenarios such as CAN bus disconnection, instantaneous sensor disconnection, or data acquisition loss: Short-term, small number of missing data points: When the number of consecutive missing data points is less than the preset threshold (e.g., consecutive missing data points ≤ 3 control cycles, i.e. ≤ 300ms), the data is filled by linear interpolation of the effective values ​​before and after the time intervals to maintain the continuity of the data.

[0052] Transient missing data: When an isolated missing data point occurs at a certain moment, the valid value from the previous moment is used directly for forward filling.

[0053] Long-term continuous missing data: When the number of consecutive missing data points exceeds the preset threshold (e.g., consecutive missing data > 10 control cycles, i.e. > 1000ms), it is determined that the communication is severely interrupted. At this time, the entire segment of consecutive missing data is truncated and discarded without any forgery or padding, so as to avoid introducing false time series trends.

[0054] Feature normalization, in particular, aims to eliminate the impact of dimensional differences between features and ensure the network can learn each input signal equally. It employs a minimum-maximum normalization method to compress all N-dimensional input features and M-dimensional output labels into the [0,1] interval, thereby improving the stability and convergence efficiency of model training. The transformation formula is as follows:

[0055] in, This is the original collected data. This is the global minimum value of this feature. This is the global maximum value of the feature. This is the result after normalization (the value is fixed as [0,1]).

[0056] Among them, the standard signal refers to the original data of the running signal that has undergone preprocessing operations such as filtering and calibration to obtain a more stable and easier-to-analyze signal form.

[0057] The window length refers to the number of historical data points included in each sample when constructing the input data for a time series prediction model; that is, the number of control periods covered by the sliding window on the time axis. For example, if data from 10 consecutive historical moments are taken as a single sample input, the window length is 10. This parameter directly determines the span of historical information that the model can "review" in each prediction and is a key hyperparameter affecting the model's ability to capture time series dependencies. It is usually designed in conjunction with a preset time domain (prediction output length).

[0058] Sliding frame assembly refers to the process of using a sliding window to continuously extract fixed-length historical data segments along the time axis with a fixed step size (usually step=1) when constructing time-series input data, and combining each segment into an independent "frame" as an input sample for the model. This method makes full use of the continuity of time-series information and maximizes the density of training samples by setting the step size to 1.

[0059] Historical data refers to the set of preprocessed and standardized multidimensional feature vectors from multiple consecutive control cycles tracing back from the current moment when constructing time-series input data. This data constitutes the complete input content of each sample within the sliding window and is the information foundation for the model to perform time-series predictions.

[0060] Thus, this implementation eliminates the sampling period differences and timing misalignments between multi-source signals through a series of preprocessing steps, repairs data defects caused by communication interruptions, removes invalid values ​​beyond physical limits and jump spikes caused by electromagnetic interference, and balances various feature scales through normalization. This provides a reliable data foundation for deep learning modeling, fault diagnosis, and strategy optimization, effectively improving the convergence speed and prediction accuracy of model training. At the same time, by using a preset window length for sliding frame grouping, continuous and equally spaced standardized time series data can be constructed into a sliding time series sample set of fixed time length. While preserving the original time series dependencies, the training sample size is expanded, thereby improving the accuracy and generalization ability of the time series prediction neural network model in predicting the future time series of the target quantity.

[0061] In one embodiment, based on a preset time-series prediction neural network model, and according to the time-series input data, the system outputs a time-series prediction result for the future preset time domain of the heat pump system, including: Input the time-series input data into a preset time-series prediction neural network model to predict the state change trend of the heat pump system within a preset time domain in the future; Based on the trend of state changes, the system outputs the time-series prediction results for the future preset time domain of the heat pump system.

[0062] Among them, the state change trend refers to the evolution direction and dynamic change law of each predicted target quantity of the heat pump system over time in the future preset time domain after the time series input data is input into the trained time series prediction neural network model.

[0063] Furthermore, if the time series prediction neural network is an LSTM (Long Short-Term Memory) neural network, the specific steps for its online time series prediction are as follows: Step 1: Based on the input signal characteristics (i.e., standard signal) and the output actuator, set the sliding window parameters, continuously slide along the time axis with a fixed step size to generate the entire dataset, and complete the dataset partitioning, including: Window length Tin=10: Use 10 consecutive historical time points as model input; Prediction step size Tout=8: Predicts the next 8 consecutive time steps; Input feature dimension: 38 (38-way input features); Output target dimension: 6 (6-channel temperature); Frame assembly logic (sliding step = 1); Training data set: 70% (the first 70% of time-series data, used for backpropagation to update weights); Validation set: 20% (middle section, adjusting parameters to observe overfitting and MSE convergence); Test data set: 10% (independent data at the end, used for final model generalization performance evaluation); Network hierarchical structure design.

[0064] Step 2, as follows Figure 2 As shown, the number of independently coupled causal relationships within the statistical system includes: Vehicle driving status Ambient temperature Air conditioning temperature and fan speed settings Evaporator outlet air temperature; Vehicle driving status Ambient temperature Air conditioning temperature and fan speed settings Warm air outlet temperature; Vehicle driving status Ambient temperature Charging and discharging current Cell temperature / temperature difference; Vehicle driving status Ambient temperature Speed Heat generation from electric motor Motor water temperature WT3; compressor speed Refrigerant pressure Chiller outlet water temperature WT4 Cell temperature / temperature difference; compressor speed Refrigerant pressure Refrigerant temperature Evaporator temperature; compressor speed Refrigerant pressure Refrigerant temperature LCC outlet water temperature WT5 Warm air outlet temperature; Multi-way directional valve (nine-way + four-way) Waterway flow direction Waterway water temperature WT1 / WT2 / WT3 / WT4 / WT5; Front fan duty cycle Ambient temperature Speed Motor water temperature WT3; BCPF pump speed Battery inlet / outlet water temperature WT1 / WT2 Battery temperature rise / temperature difference; HWPC heater pump Water flow velocity Water temperature in the water system: WT5; EDCP speed of motor water pump Water flow velocity Water temperature in the water circuit is WT3; Step 3: Based on the number of independent coupled causal relationships within the system, 12 sets of independent physical laws are obtained. Each law is encoded using 8 neurons. The first layer of the neural network is designed with 96 neurons to accommodate all coupled laws; the second layer is set to 48 neurons according to the principle of halving the number of neurons per layer. The neural network hierarchy is as follows: Input layer: [Number of samples, 10 window length, 38 input features] First LSTM layer: 96 neurons, Dropout=0.2 (randomly masking 20% ​​of neurons to prevent overfitting) Second LSTM layer: 48 neurons Output layer: Outputs the timing sequence for the next 8 steps, and predicts the target dimensions in 6 channels (average battery temperature, maximum battery temperature difference, chiller outlet water temperature, warm air outlet temperature, evaporator outlet temperature, and motor circuit water temperature). Step 4: Training and Lightweight Deployment The dataset covers all operating conditions from -35℃ to 55℃, including high and low temperatures, fast charging, hill climbing, idling, high speed, heating, and cooling. The loss function uses the mean squared error (MSE) and the Adam adaptive optimizer. The basic mathematical formula for MSE is as follows: Single sample, single prediction point:

[0065] This is the actual label value (the true temperature value measured by the actual vehicle sensor). The predicted value is from the LSTM network. The total number of prediction points participating in the summation. To account for prediction errors, squaring ensures that positive and negative errors do not cancel each other out.

[0066] Total number of prediction points per sample: N = 8 × 6 = 48 (8 future time steps × 6 temperature variables) Single sample loss formula:

[0067] This is the t-th prediction step (t = 1 - 8 time steps). The k-th predicted temperature (k=1-6, corresponding to the average battery temperature, temperature difference, chiller water temperature, heater temperature, evaporator temperature, and motor water temperature).

[0068] Total loss for a batch containing B samples:

[0069] Step Six: MSE drives the automatic division of labor among the 96 neurons: Forward propagation: Input 10 steps of historical data LSTM calculated 48 temperature prediction values; Calculate the total MSE loss to measure the overall prediction error. Backpropagation (BP) + Adam optimizer: calculate gradients layer by layer from the output layer back, and fine-tune the weights inside all LSTM neurons little by little; It can reduce the weight change of MSE The gradient descent direction is used to fine-tune the weights; Increase MSE adjustment Reverse correction, suppression; After tens of thousands of iterations: Regarding "compressor speed" The weight of neurons sensitive to "cell temperature / temperature difference" is continuously increased, effectively reducing the prediction error corresponding to this pattern; For "BCPF water pump" Another batch of neurons, sensitive to battery temperature, experienced simultaneous weight enhancement. The weights of irrelevant inputs are continuously reduced to near 0; Ultimately, the 80 neurons spontaneously formed different specializations, and the overall MSE continued to converge and decrease, making the predictions more and more accurate.

[0070] In summary, MSE (mean squared error) acts as a "scorer" for neural networks. For example, the larger the loss value it outputs, the more severe the deviation between the predicted result and the actual value. During training, the neural network aims to minimize this score, continuously adjusting its internal weights and biases and other learnable parameters through backpropagation. Each iteration updates the network parameters in the direction of loss reduction. In this process, gradient information is distributed differentially across different neurons, and each neuron gradually develops the ability to capture specific input features or physical laws. For example, some neurons are sensitive to temperature changes, while others are sensitive to pressure fluctuations, thus achieving a distributed representation and efficient learning of the complex coupling laws of the heat pump system as a whole.

[0071] Step 5: After training, pruning and weight reduction are performed, and the system is deployed to the vehicle's VCU thermal management controller. The single inference time is <10ms, meeting the 100Hz vehicle control cycle. The system predicts temperature hysteresis trends, providing a future time-domain prediction model for the MPC controller.

[0072] Thus, this implementation method uses a time-series predictive neural network to accurately predict the future state change trend of the heat pump system, upgrading the control decision from "passive response" to "active prediction". This improves the temperature control accuracy and energy efficiency while enhancing the system's operational safety and providing reliable forward-looking input support for MPC.

[0073] In one embodiment, before obtaining the target control sequence by performing MPC multi-objective rolling optimization solution under preset constraints based on the time series prediction results, the method further includes: Determine the current operating mode of the heat pump system based on standard signals; Set the preset initial values ​​for MPC multi-objective rolling solution based on the target working condition mode.

[0074] Among them, the operating mode refers to the system's dominant control objective and operating strategy determined by the new energy vehicle's heat pump system based on a comprehensive judgment of the current operating scenario, considering factors such as vehicle driving status, environmental conditions, passenger comfort needs, and battery thermal management requirements. Examples include: cabin cooling mode, battery cooling mode, cabin cooling + battery cooling mode, cabin dehumidification + battery cooling mode, cabin dehumidification + battery heating mode, cabin heat pump heating mode, battery heat pump heating mode, and cabin heat pump heating + battery heat pump heating mode.

[0075] The preset initial value refers to the starting point of the iterative calculation of the MPC multi-objective rolling optimization solver in the current control cycle. It includes the actual state of the system at the current moment, the actual output of each actuator at the previous moment, and the baseline predicted trajectory based on the unchanged current control quantity. The preset initial value is a key input parameter to ensure rapid convergence of the optimization solver and meet the real-time requirements of the vehicle controller. For example: electric compressor: target speed; BEXV (Chiller heat exchanger expansion valve): initial opening; EEXV (EVAP evaporator expansion valve): initial opening; AEXV (heat pump gas-side electronic expansion valve): initial opening; battery water pump BCPF: initial speed; heater water pump HCWP: initial speed; motor loop water pump EDCP: initial speed; nine-way reversing valve mode opening: initial opening; four-way reversing proportional valve opening: initial opening; fan operating speed: initial duty cycle.

[0076] Thus, by providing physically feasible and approximate optimal solution-approximate iterative initial values ​​for MPC multi-objective rolling optimization, this implementation can significantly accelerate solution convergence to ensure the real-time performance of the vehicle controller, improve solution stability, ensure smooth actuator action, and achieve efficient coupling between timing prediction neural network prediction and MPC optimization, thereby comprehensively improving the response speed, control quality, and robustness of the control system under dynamic operating conditions.

[0077] In one embodiment, based on the time-series prediction results, MPC multi-objective rolling optimization is performed under preset constraints to obtain the target control sequence, including: Based on the preset initial values ​​and timing prediction results, the multi-objective solution is performed using the MPC multi-objective rolling optimization algorithm under preset constraints, and the target control sequence of each actuator in the control time domain is obtained.

[0078] Among them, the control time domain refers to the future time range covered by the control commands calculated and output by MPC in a single optimization solution.

[0079] Furthermore, the specific steps of the MPC solution logic include: Step 1: Obtain the current system state and the future predicted temperature sequence of the time-series prediction neural network. For example, collect the current ambient temperature, vehicle status, battery temperature, cabin temperature, water temperature of each circuit, current compressor speed, valve opening degree, and speed of the three water pumps in real time. And call the upper-level time-series prediction neural network model to obtain the predicted temperature sequence of the next Np steps: average battery temperature, temperature difference, chiller water temperature, heater temperature, evaporator temperature, motor water temperature, etc.

[0080] Step 2: Define the incremental sequence of the controlled object and control the time domain length. Solve for a whole set of future actions:

[0081] Each All are 8-dimensional increments, corresponding to the changes in compressor speed, the changes in the opening of the three expansion valves, the changes in the speed of the three water pumps, and the changes in the Fan duty cycle.

[0082] Step 3: Input the system's dynamic power consumption calculation physical model to predict the total future power consumption.

[0083] Step-by-step deduction of the compressor power Pcomp in the future Np steps: Adjust compressor speed Refrigerant circulation flow rate and pressure changes The compressor power Pcomp changes; Adjust BEXV / EEXV / AEXV opening Changes in refrigerant throttling intensity alter the heat exchange capacity of heat exchangers and evaporators. The compressor power Pcomp changes; Adjust the speed of BCPF / HCWP / EDCP water pumps Changes in water flow alter the heat exchange rates of the battery, heater, and motor circuits. The compressor power Pcomp changes; Adjust the ratio of the nine-way valve and the four-way valve The water flow direction changes, and the loop coupling relationship changes. The compressor power Pcomp changes; Ultimately, each step in the future is deduced. , , , Substitute into the cost function to calculate the total loss .

[0084] in, The cost function is

[0085] First item: Predict the sum of multi-objective penalties in the time domain (Np-step future prediction) This is a penalty term representing the square of the deviation between the battery's average temperature and the target value; the larger the deviation, the exponentially greater the cost contribution. MPC actively adjusts the commands of each actuator through optimization, driving the battery's average temperature to approach the target value. This represents the average battery temperature. To set the target temperature.

[0086] A penalty term based on the square of the maximum temperature difference between individual battery cells is used to suppress the temperature dispersion trend within the battery pack. MPC actively reduces the temperature difference between cells by optimizing the allocation of cooling / heating flow in each branch, thereby improving the temperature uniformity of the battery pack, enhancing charge / discharge balance, and extending cycle life. This represents the maximum temperature difference of a single battery cell.

[0087] This is a penalty item for cabin temperature tracking, used to quantify the deviation between the actual temperature and the set temperature. The MPC (Manual Control Program) actively adjusts the cabin temperature towards the set value by coordinating the actions of actuators such as the compressor, blower, and dampers, ensuring the accurate achievement of the air conditioning's cooling / heating targets. This refers to the actual cabin temperature. Set the temperature for the occupants.

[0088] As a real-time power consumption penalty for the compressor, the MPC tends to choose a control strategy with a lower compressor speed to achieve the same temperature control effect, thereby reducing system energy consumption and improving the overall energy efficiency ratio (COP) of the heat pump system. This represents the real-time electrical power consumption of the compressor at the current speed.

[0089] The above represents a weighted compromise between four conflicting objectives: temperature control accuracy, battery consistency, comfort, and energy consumption.

[0090] Second item: Controlled incremental smoothing penalty (Nc controls the time domain)

[0091] (The change in the actuator's action at the current moment relative to the previous moment) It is an 8-dimensional vector, with the following complete definition: =[comp speed, BEXV opening, EEXV opening, AEXV opening, BCPF speed, HCWP speed, EDCP speed, Fan duty cycle]T weight matrix A diagonal matrix, with each actuator corresponding to a smoothing coefficient; the larger the coefficient, the less allowable the actuator is to undergo large abrupt changes, suppressing valve jumps, frequent pump starts and stops, and sudden increases and decreases in compressor speed, thus avoiding water hammer in pipelines, refrigerant pressure shocks, and large temperature fluctuations.

[0092] Step 4: Add global hard constraints (solve the boundary to prevent over-range).

[0093] The upper and lower bounds that must be satisfied to solve the problem include: compressor: The upper limit of the amplitude of a single change in rotational speed; among which, This is the minimum permissible speed of the compressor. The target compressor speed in the k-th control cycle, This refers to the compressor's maximum permissible speed. Expansion valve: opening degree 0%~100%, maximum adjustment step of single opening degree; Three water pumps: each with its own upper and lower speed limits and speed regulation rate limits; Nine-way valve and four-way valve: mutually exclusive positions and time-delay constraint for reversing action; System safety constraints: refrigerant high and low pressure extremes, lower limit of superheat, safe range of battery temperature, upper and lower limits of water temperature in different circuits.

[0094] Step 5: Numerical optimization to minimize the total cost Solve for the optimal control increment Multi-objective processing: This formula uses weighted sum scalarization, with four objectives handled using... - Weighted merging into a single objective function; Solver solution: Total cost function merging Standard QP quadratic programming form; Absolute upper and lower limit constraints for actuators; Single-step adjustment rate constraint (incremental limit); System safety constraints (temperature, pressure, superheat); Output the optimal sequence:

[0095] Step Six: Extract the current output and send the target values ​​for the compressor, water pump, valve, and fan. Current optimal increment ; By superimposing the executor's actual value from the previous moment, we obtain the current target value:

[0096] Among them, the core rule of MPC rolling control is: only the first step is executed, the subsequent sequence is discarded, and the solution is recalculated in the next cycle.

[0097] Step 7: Output the final target in 8 channels: Electric compressor: target speed; BEXV (Chiller heat exchanger expansion valve): Target opening degree; EEXV (EVAP evaporator expansion valve): Target opening degree; AEXV (Heat Pump Gas-Side Electronic Expansion Valve): Target opening degree; Battery-powered water pump BCPF: Target speed; HCWP (Heat Pump HCWP): Target Speed; Motor-driven water pump EDCP: Target speed; Fan operating speed: target duty cycle; Entering the next control cycle, the state is re-acquired, the time-series predictive neural network is used for prediction, and the MPC is solved in a rolling manner, running in a closed loop.

[0098] Thus, this implementation method achieves the optimal trade-off between temperature control accuracy, energy economy, and safety margin by solving the multi-objective cost function online, while satisfying the physical constraints of the system. It combines the prediction of the time-series predictive neural network to achieve feedforward control, and improves the robustness of the system to disturbances and model biases through rolling optimization and feedback correction. Ultimately, it maximizes the energy efficiency of the heat pump system while ensuring battery safety and cabin comfort.

[0099] In one embodiment, outputting the target control sequence to the corresponding actuator to execute the corresponding cooperative action instruction includes: Based on the preset data arrangement rules of the target control sequence, determine the position index number of each value in the control quantity sequence at the current moment; Based on the correspondence between position index number and actuator type, the target actuator category and its control parameters matched for each control variable are obtained; Based on the target actuator type, the corresponding actuator driver interface is called to convert the control quantity into a drive signal and output it, thus obtaining the coordinated action command of multiple actuators in the heat pump system.

[0100] Among them, the data arrangement rule refers to the fixed mapping convention between the value of each control variable and a specific actuator type and control parameter in the target control sequence.

[0101] The position index number is a unique integer number, starting from 0 or 1, assigned to each control value in the data arrangement rule. It is used to quickly locate the corresponding position in the arrangement rule for the currently processed value during the parsing process. For example, the MPC solver outputs the target control sequence in the current control cycle. Based on the preset data arrangement rule, the system determines the correspondence between position index numbers 0 to 6: index number 0 corresponds to the compressor target speed, index number 1 to the battery water pump target speed, index number 2 to the motor water pump target speed, index number 3 to the main circuit electronic expansion valve target opening, index number 4 to the battery cooling circuit electronic expansion valve target opening, index number 5 to the three-way water valve target opening, and index number 6 to the condenser fan target duty cycle. After parsing, the target values ​​are converted into drive signals through the actuator drive interfaces and output synchronously, forming the coordinated action instructions of the heat pump system in the current control cycle.

[0102] Thus, this implementation method, through a mapping mechanism of preset data arrangement rules and position index numbers, accurately and efficiently transforms the target control sequence output by MPC into drive signals that can be recognized by each actuator, realizing the generation of collaborative action instructions of multiple actuators under a unified time base, ensuring unambiguous transmission between control decisions and physical execution, thereby guaranteeing the real-time performance, determinism and execution consistency of the entire thermal management system under the multi-objective optimization framework.

[0103] In one embodiment, the method further includes: Based on the actual system response data collected after the control input is output to the corresponding actuator, the real-time measured value of the current status is obtained. Based on the real-time measured values ​​and the preset "first-in-first-out" update rule of the sliding window, the update result of the sliding window is obtained; Based on the update results, the historical data sequence used in the next control time domain is obtained.

[0104] Among them, the real-time measured value refers to the actual physical value of the key system variables obtained by each sensor in the current control cycle after preprocessing.

[0105] The "first-in, first-out" update rule of the sliding window refers to a data update mechanism that, when constructing time-series input data, whenever new control cycle data arrives, the sliding window will move the oldest time step data out of the window and move the latest time step data into the window, thereby maintaining a constant window length.

[0106] Thus, this implementation method uses a first-in-first-out (FIFO) update rule of "removing the oldest and moving in the newest" to ensure that the sliding window always provides the time-series prediction neural network model with a continuous and up-to-date historical state sequence while maintaining a fixed time span. This ensures that online prediction and MPC rolling optimization can respond in real time based on the current system state.

[0107] In one embodiment, the method further includes: The rationality of the prediction results of the time series prediction neural network is diagnosed. If the diagnosis result is abnormal, the first abnormal monitoring signal is generated. The convergence status of the MPC multi-objective optimization solution is monitored. If non-convergence is detected, a second abnormal monitoring signal is generated. When any of the above monitoring signals is triggered, the control system automatically switches from the current timing prediction-MPC cooperative control mode to the preset PID fallback control mode.

[0108] The first abnormal monitoring signal refers to the early warning signal generated by the rationality diagnosis module after the time-series prediction neural network model completes the prediction of the state within the preset time domain, and the prediction result is verified by the rationality diagnosis module. When it is determined that there is an abnormality in the prediction result, it is used to mark the abnormal state, trigger protection actions, or switch control strategies.

[0109] The second abnormal monitoring signal refers to the emergency warning signal generated by the convergence state monitoring module during the MPC multi-objective rolling optimization process, which monitors the iterative process of the solver in real time and triggers the highest level of security protection action when the optimization solution is detected to be non-convergent.

[0110] Thus, this implementation establishes a dual-mode redundant control architecture that prioritizes predictive optimization and supplements it with PID control. When the time-series predictive neural network fails to predict or the MPC solution fails, it automatically and smoothly switches to a conservative PID control strategy. This ensures that the thermal management system always has effective control output under all operating conditions, avoiding the risk of actuator malfunction or loss of control due to the failure of the intelligent control module. As a result, it improves energy efficiency and comfort under normal operating conditions while ensuring system safety and control continuity under extreme operating conditions.

[0111] Based on the same inventive concept as the foregoing embodiments, the foregoing embodiments will be described in detail below through a specific application scenario, such as... Figure 2 As shown, 1. Refrigerant heat pump system circulation Electric compressor exhaust via PT1 pressure and temperature sensor Two-way LCC agent-side heat exchangers with AEXV bypass valve for pressure relief and reflux. Then split the flow into two paths: EEXV and BEXV electronic expansion valves. After the flow is split, the two streams pass through the EVAP and Chiller agent-side heat exchangers, respectively. After the three signals converge, they pass through the PT2 pressure and temperature sensor. The compressor's suction port forms a refrigerant circuit for heat transfer, providing a highly efficient refrigerant-side circulation system for refrigeration and heating cycles.

[0112] 2. Battery water circuit Nine-way valve corresponding interface Battery-Powered Water Pump (BCPF) WT2 water temperature sensor Battery pack WT1 water temperature sensor Nine-way valve corresponding interface; 3.chiller water circuit Nine-way valve corresponding interface Chiller heat exchanger water side WT4 water temperature sensor Nine-way valve corresponding interface; 4. Heating system and LCC water circuit Nine-way valve corresponding interface Four-way valve directional valve corresponding interface HCWP heater pump LCC heat exchanger water side WT5 water temperature sensor The flow is split into two paths to the corresponding interfaces of the Heater core and the nine-way valve, with the Heater core circuit... Four-way valve reversing valve corresponding interface; 5. Corresponding interface of the nine-way valve in the motor-controlled water circuit. Front-end heatsink radiator / fan and bypass circuit EDCP motor water pump R-EDS cooling assembly for drive motor WT3 water temperature sensor Nine-way valve corresponding interface; 6. Core commutation actuator Nine-way reversing valve: 9 water circuit interfaces to realize the connection and disconnection between battery circuit, heating circuit, motor circuit, chiiller heat exchange circuit and radiator circuit. It is the core component for multi-mode switching.

[0113] Four-way proportional reversing valve: It has three built-in opening ratio modes to switch the flow direction of the heating circuit and the heat dissipation flow direction of the motor circuit. Through proportional adjustment, the heat pump system can switch between cooling and heating cycle modes.

[0114] Based on the same inventive concept as the foregoing embodiments, the foregoing embodiments will be described in detail below with specific examples, wherein the temporal prediction neural network model is illustrated using the LSTM neural network model as an example.

[0115] This example uses the cabin cooling mode as an example to illustrate the operating conditions.

[0116] Valve reversing configuration When the nine-way valve is switched to mode one, the battery circuit is connected to the Chiller circuit, and the motor circuit is connected to the LCC circuit; when the four-way valve PFWV is switched to mode A, that is, interfaces 2-3 are connected, and other interfaces are not connected; after the refrigerant is discharged from the compressor and releases heat through the LCC, it enters the evaporator of the air conditioning unit after being throttled by the EEXV to absorb heat, thereby achieving cabin cooling.

[0117] LSTM prediction task Anticipate the rising trends of evaporator temperature and refrigerant pressure, and predict changes in refrigeration load.

[0118] MPC actuator cooperative control strategy Compressor: Dynamically adjusts speed to match refrigeration load, suppresses excessively high pressure, and improves refrigeration COP; EEXV: Main throttle valve, which precisely adjusts its opening degree according to the evaporator superheat; BEXV: Turn off standby; AEXV: Disable standby; BCPF (Battery Pump): Turn off standby; HCWP (Heat Pump): Adjustable as needed, ensuring that the heat released by the LCC circulates through the water circuit to the motor circuit and into the air; EDCP (Electric Motor Pump): Adjustable as needed; when the LCC and motor overheat, the front-end radiator is activated for heat dissipation. Fan: Adjustable as needed to ensure that the heat released by the LCC can be quickly dissipated through the heatsink; Control objectives: cabin temperature quickly follows the set value, system energy consumption is minimized, and refrigerant pressure is stable without fluctuations.

[0119] This example uses the cabin heat pump heating mode (low winter temperatures, heating demand) as an example for illustration. Valve reversing configuration When the nine-way valve is switched to mode four, the motor circuit is interconnected with the Chiller circuit, and the battery circuit is interconnected with the LCC circuit; when the four-way valve PFWV is switched to mode B, that is, interfaces 2-4 are connected, and other interfaces are not connected; the heat pump operates in reverse circulation, and the LCC acts as a condenser to release heat. That is, after the refrigerant is discharged from the compressor and releases heat through the LCC, it enters the Chiller heat exchanger after being throttled by the BEXV and absorbs heat through the motor water circuit to achieve cabin heating.

[0120] LSTM prediction task Predict the temperature rise slope of the heating air outlet and the heating air circuit water temperature to prevent temperature overshoot and fluctuations, and predict changes in heating load.

[0121] MPC actuator cooperative control strategy Compressor: Dynamically adjusts speed to match heating load, suppresses excessively low pressure, and improves heating COP; EEXV: Turn off standby; BEXV: Adjusts as needed to ensure the chiller can absorb heat from the motor water circuit; AEXV: Adjusts as needed to ensure that the compressor suction superheat and low pressure are within the compressor's required range; BCPF (Battery Pump): Turn off standby HCWP (Heat Pump): Adjusts as needed to ensure that the heat released by the LCC circulates through the water circuit to the heating circuit and into the vehicle. EDCP (Electric Motor Pump): Adjustable as needed to ensure that the Chiller heat exchanger can absorb waste heat from the motor and air from the motor circuit; Fan: Adjustable as needed to ensure the heat pump can absorb enough heat from the outside through the radiator; Control objectives: Stable cabin temperature rise, maximize heat pump heating efficiency, and reduce overall vehicle heating power consumption.

[0122] This example uses the battery active cooling mode (fast charging / high load discharging, battery over-temperature > 40℃) as an example for illustration. Valve reversing configuration When the nine-way valve is switched to mode one, the battery circuit is connected to the Chiller circuit, and the motor circuit is connected to the LCC circuit; when the four-way valve PFWV is switched to mode A, that is, interfaces 2-3 are connected, and other interfaces are not connected; after the refrigerant is discharged from the compressor and releases heat through the LCC, it enters the Chiller heat exchanger after being throttled by the BEXV to absorb heat from the battery water circuit, thereby achieving battery cooling.

[0123] LSTM prediction task By predicting the future rate of temperature rise in the battery and the changes in temperature difference between individual battery cells, adjustments can be made in advance to suppress temperature spikes.

[0124] MPC actuator cooperative control strategy Compressor: Dynamically adjusts speed to match refrigeration load, suppresses excessively high pressure, and improves refrigeration COP; EEXV: Turn off standby; BEXV: Adjusts on demand, controls Chiller overheating, and precisely controls battery cooling intensity; AEXV: Disable standby; BCPF (Battery Pump): Adjusts as needed, linearly speed-regulated to match cooling flow, suppressing battery temperature differences; HCWP (Heat Pump): Adjustable as needed, ensuring that the heat released by the LCC circulates through the water circuit to the motor circuit and into the air; EDCP (Electric Motor Pump): Adjustable as needed; when the LCC and motor overheat, the front-end radiator is activated for heat dissipation. Fan: Adjustable as needed to ensure that the heat released by the LCC can be quickly dissipated through the heatsink; Control objectives: The battery temperature should drop rapidly to the optimal range of 25-35℃, the temperature difference between individual cells should be stably controlled within 5℃, the temperature fluctuation should be <±0.5℃, the system energy consumption should be minimized, and the refrigerant pressure should be stable without vibration.

[0125] This example uses the battery low-temperature preheating mode (battery temperature <10℃, winter charging / driving preheating) as an example for illustration. Valve reversing configuration When the nine-way valve is switched to mode four, the motor circuit is interconnected with the Chiller circuit, and the battery circuit is interconnected with the LCC circuit; when the four-way valve PFWV is switched to mode B, that is, interfaces 2-3 are connected, and other interfaces are not connected; the heat pump operates in reverse circulation, and the LCC acts as a condenser to release heat. That is, after the refrigerant is discharged from the compressor and releases heat through the LCC, it enters the Chiller heat exchanger after being throttled by the BEXV and absorbs heat through the motor water circuit to achieve battery heating.

[0126] LSTM prediction task Predict the battery's temperature rise rate to prevent localized overheating and a widening temperature difference between the two ends.

[0127] MPC actuator cooperative control strategy Compressor: Dynamically adjusts speed to match heating load, suppresses excessively low pressure, and improves heating COP; EEXV: Turn off standby; BEXV: Adjusts as needed to ensure the chiller can absorb heat from the motor water circuit; AEXV: Adjusts as needed to ensure that the compressor suction superheat and low pressure are within the compressor's required range; BCPF (Battery Pump): Adjusts as needed, linearly speed-regulated to match cooling flow, suppressing battery temperature differences; HCWP (Heat Pump): Adjusts as needed to ensure that the heat released by the LCC circulates through the water circuit to the heating circuit and into the vehicle. EDCP (Electric Motor Pump): Adjustable as needed to ensure that the Chiller heat exchanger can absorb waste heat from the motor and air from the motor circuit; Fan: Adjustable as needed to ensure the heat pump can absorb enough heat from the outside through the radiator; Control objectives: The battery temperature rises smoothly to around 25°C, with uniform temperature rise and no localized high temperatures, while taking into account the utilization rate of waste heat, maximizing the heating efficiency of the heat pump, and reducing the overall vehicle heating power consumption.

[0128] This example uses the motor and electronic control high-temperature heat dissipation mode as an example for illustration. When the temperature rise of the motor and ADCU exceeds the standard, the nine-way valve switches to radiator mode, and the EDCP water pump drives the coolant to flow through the front radiator for forced air cooling; the other circuits go into hibernation as needed, and the MPC adjusts the water pump speed and fan duty cycle independently to control the motor temperature without interfering with the operation of the battery and air conditioning circuits.

[0129] EDCP (Electric Motor Pump): Adjustable on demand, linear speed regulation matches cooling flow, and improves the heat exchange capacity of the motor water circuit radiator; Fan: Adjustable as needed, it can quickly dissipate heat from the motor circuit into the air; Control objective: To balance the water temperature in the motor circuit and the motor efficiency, thereby maximizing heat dissipation efficiency and reducing the power consumption of low-voltage components in the vehicle.

[0130] In summary, the method provided in the above embodiments collects vehicle operating signals and preprocesses them to obtain time-series input data. Then, based on a preset time-series prediction neural network model, it outputs the time-series prediction results of the heat pump system within a preset time domain, according to the time-series input data. Based on the time-series prediction results, it performs MPC multi-objective rolling optimization under preset constraints to obtain the target control sequence. Finally, based on the target control sequence, it controls the corresponding actuators to execute corresponding cooperative action commands. Thus, through the deep integration of time-series prediction and MPC rolling optimization, the problems of strong multi-loop coupling, large hysteresis, and nonlinear disturbances in the heat pump system are effectively solved. Under the premise of satisfying the system's physical constraints, it achieves comprehensive technical effects such as improved temperature control accuracy, optimized heat pump energy efficiency, reduced vehicle energy consumption, and extended battery life.

[0131] Based on the same inventive concept as the foregoing embodiments, this application provides a computing device, such as... Figure 3 As shown, the device includes: a processor 310 and a memory 311 storing a computer program; wherein, Figure 3 The processor 310 shown in the diagram does not indicate that there is only one processor 310, but only indicates the positional relationship of the processor 310 relative to other devices. In practical applications, there can be one or more processors 310; similarly, Figure 3 The memory 311 shown in the diagram has the same meaning, that is, it is only used to indicate the positional relationship of memory 311 relative to other devices. In practical applications, there can be one or more memories 311. When the processor 310 runs the computer program, it implements the control method of the heat pump system applied to the above-mentioned device.

[0132] The device may also include at least one network interface 312. The various components of the device are coupled together via a bus system 313. It is understood that the bus system 313 is used to implement communication between these components. In addition to a data bus, the bus system 313 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in... Figure 3 The general designated all buses as Bus System 313.

[0133] The memory 311 can be volatile memory or non-volatile memory, or both. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), ferromagnetic random access memory (FRAM), flash memory, magnetic surface memory, optical disc, or compact disc read-only memory (CD-ROM); magnetic surface memory can be disk storage or magnetic tape storage. Volatile memory can be random access memory (RAM), used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Synchronous Static Random Access Memory (SSRAM), Dynamic Random Access Memory (DRAM), Synchronous Dynamic Random Access Memory (SDRAM), Double Data Rate Synchronous Dynamic Random Access Memory (DDRSDRAM), Enhanced Synchronous Dynamic Random Access Memory (ESDRAM), SyncLink Dynamic Random Access Memory (SLDRAM), and Direct Rambus Random Access Memory (DRRAM).The memory 311 described in the embodiments of this application is intended to include, but is not limited to, these and any other suitable types of memory.

[0134] The memory 311 in this embodiment is used to store various types of data to support the operation of the device. Examples of this data include: any computer programs used to operate on the device, such as operating systems and applications; contact data; phonebook data; messages; pictures; videos, etc. The operating system includes various system programs, such as the framework layer, core library layer, driver layer, etc., used to implement various basic services and handle hardware-based tasks. Applications can include various applications, such as media players, browsers, etc., used to implement various application services. Here, the program implementing the method of this embodiment can be included in the application.

[0135] Based on the same inventive concept as the foregoing embodiments, this embodiment also provides a computer-readable storage medium storing a computer program. The computer-readable storage medium can be a magnetic random access memory (FRAM), a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a flash memory, a magnetic surface memory, an optical disc, or a compact disc read-only memory (CD-ROM), etc.; it can also be various devices including one or any combination of the above-mentioned memories, such as mobile phones, computers, tablet devices, personal digital assistants, etc. When the computer program stored in the computer-readable storage medium is executed by a processor, it implements the control method of the heat pump system described above. For the specific steps implemented when the computer program is executed by the processor, please refer to [link to relevant documentation]. Figure 1 The description of the illustrated embodiments will not be repeated here.

[0136] 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.

[0137] In this document, the terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, which includes not only the elements listed but also other elements not expressly listed.

[0138] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A control method for a heat pump system, characterized in that, include: Collect vehicle operating signals; The operating signals are preprocessed to obtain timing input data; Based on a preset time-series prediction neural network model, the system outputs a time-series prediction result for the heat pump system within a preset time domain, according to the time-series input data. Based on the time series prediction results, MPC multi-objective rolling optimization is performed under preset constraints to obtain the target control sequence. Based on the target control sequence, the corresponding actuator is controlled to execute the corresponding cooperative action instructions.

2. The method as described in claim 1, characterized in that, The step of preprocessing the operating signal to obtain timing input data includes: The raw data of the operating signal is subjected to time alignment, filtering and noise reduction, missing value interpolation and normalization to obtain the standard signal of the heat pump system. The standard signal and its corresponding timestamp are framed by sliding within a preset window length to obtain the timing input data; wherein, the timing input data includes the current time and historical data from multiple past times.

3. The method as described in claim 1, characterized in that, The preset time-series prediction neural network model outputs a time-series prediction result for the heat pump system within a preset time domain based on the time-series input data, including: The time-series input data is input into a preset time-series prediction neural network model to predict the state change trend of the heat pump system within a preset time domain in the future. Based on the stated trend of state change, the time-series prediction results of the heat pump system within the future preset time domain are output.

4. The method as described in claim 1, characterized in that, Before obtaining the target control sequence by performing MPC multi-objective rolling optimization solution under preset constraints based on the time series prediction results, the method further includes: Based on the standard signal, determine the current operating mode of the heat pump system; Set the preset initial values ​​for MPC multi-objective rolling solution based on the target working condition mode.

5. The method as described in claim 1, wherein the step of performing MPC multi-objective rolling optimization solution under preset constraints based on the time series prediction result to obtain the target control sequence includes: Based on the preset initial value and the timing prediction result, under preset constraints, the multi-objective solution is performed using the MPC multi-objective rolling optimization algorithm to obtain the target control sequence of each actuator in the control time domain.

6. The method as described in claim 1, characterized in that, The step of outputting the target control sequence to the corresponding actuator to execute the corresponding cooperative action instruction includes: According to the preset data arrangement rules of the target control sequence, determine the position index number of each value in the control quantity sequence at the current moment; Based on the correspondence between the position index number and the actuator type, the target actuator category and its control parameters matched for each control variable are obtained; Based on the target actuator category, the corresponding actuator drive interface is invoked to convert the control quantity into a drive signal and output it, thereby obtaining the coordinated action command of multiple actuators in the heat pump system.

7. The method as described in claim 1, characterized in that, Also includes: Based on the actual system response data collected after the control quantity is output to the corresponding actuator, the real-time measured value of the current state is obtained. Based on the real-time measured values ​​and the preset "first-in-first-out" update rule of the sliding window, the update result of the sliding window is obtained; Based on the update results, the historical data sequence used in the next control time domain is derived.

8. The method as described in claim 1, characterized in that, Also includes: The rationality of the prediction results of the time series prediction neural network is diagnosed. If the diagnosis result is abnormal, the first abnormal monitoring signal is generated. The convergence status of the MPC multi-objective optimization solution is monitored. If non-convergence is detected, a second abnormal monitoring signal is generated. When any of the above monitoring signals is triggered, the control system automatically switches from the current timing prediction-MPC cooperative control mode to the preset PID fallback control mode.

9. A computing device, characterized in that, include: The processor and the memory storing a computer program implement the steps of the control method for the heat pump system according to any one of claims 1 to 8 when the processor runs the computer program.

10. A computer-readable storage medium, characterized in that, The system contains a computer program that, when executed by a processor, implements the steps of the control method for the heat pump system according to any one of claims 1 to 8.

Images