A heat pump control method and system based on status data

Abstract

The present invention relates to the technical field of heat pump heating, and particularly relates to a heat pump control method and system based on state data. The method includes: obtaining a training set; training a network prediction model using the training set; inputting the state data of the heat pump at the current moment into the network prediction model to obtain the predicted temperature at the next moment; comparing the predicted temperature with the target outlet water temperature to achieve the control of the heat pump. The solution of the present invention can accurately control the heat pump.

Description

Technical Field

[0001] The present invention relates to the technical field of heat pump heating. More specifically, the present invention relates to a heat pump control method and system based on state data. Background Art

[0002] A heat pump is a device that utilizes low-grade thermal energy in the external environment to achieve energy conversion through a compressor and a refrigerant, so as to supply hot water or hot air. The heat pump uses the refrigerant to evaporate at a low temperature to absorb environmental heat, and then discharges heat after compression, and supplies high-grade heat sources (hot water or hot air) to a building or a water system, thereby achieving purposes such as heating.

[0003] Heat pumps are commonly used in household heating, industrial production, commercial buildings, and agricultural breeding, etc. For example, in the industrial production processes of food and medicine, it is necessary to heat items at a constant temperature. At this time, as a device that can adjust the temperature, the heat pump can play a relatively positive role in production efficiency.

[0004] For example, in the related art, a Chinese patent application document with the publication number CN115560374A discloses a heating control method and system based on the processing of heat pump unit state data. It obtains a data set of the air inlet temperature of the unit and a data set of the heating capacity of the unit; generates an air inlet temperature - heating capacity change curve and generates a fitting coefficient; and inputs the real-time operating external environment index and the preset heating index into an adaptive control model to output adaptive control parameters for controlling the target heat pump unit.

[0005] However, since the data to be collected is realized through sensors, and due to the influence of the environment, the sensors themselves, and the equipment operation process during data collection, there may be noise data in the collected data; at the same time, the influence of data in different dimensions on heat pump control is different.

[0006] Therefore, how to effectively and accurately control the temperature of the heat pump during the heating process is particularly important. Summary of the Invention

[0007] The purpose of the present invention is to propose a heat pump control method and system based on state data to solve the problem that the heat pump cannot be effectively and accurately controlled in the prior art; for this purpose, the present invention provides solutions in the following two aspects.

[0008] In the first aspect, a heat pump control method based on state data provided by the present invention includes:

[0009] Obtain a training set; the training set is a plurality of state data within a historical set time sequence and the labels of each state data, and the label is the outlet water temperature at the next moment of the moment when each state data is located; the state data includes data in multiple dimensions;

[0010] Train a network prediction model using the training set;

[0011] Among them, the loss function during training is the mean squared error loss with weights; the weights are: ; is the coordination degree of the nth dimension data in the kth state data, is the noise degree of the nth dimension data in the kth state data, N is the total number of dimensions in the state data, exp( ) is the exponential function; the coordination degree characterizes the coupling relationship between the time series of any dimension and the target time series, the target time series is the time series composed of all labels, and the time series of any dimension is the time series composed of data of the same dimension; the noise degree characterizes the situation where the data of each dimension belongs to noise;

[0012] Input the state data of the heat pump at the current moment into the network prediction model to obtain the predicted temperature at the next moment; compare the predicted temperature with the target outlet water temperature to achieve the control of the heat pump.

[0013] The above solution can obtain the contribution size of the data of any dimension in the state data during the training process by analyzing the correlation relationship between each dimension and the label in multiple dimensions and the noise situation of the data of each dimension itself, and then comprehensively considering the contribution situation of the data of all dimensions in the state data to determine the importance of the corresponding state data, so as to adjust the difference between the predicted value and the label during the training process to obtain the loss value, improve the accuracy of the training of the network prediction model, and thus accurately predict the predicted temperature at the future moment for the subsequent effective control of the heat pump.

[0014] Optionally, the data of the multiple dimensions includes but is not limited to the inlet water temperature, the outlet water temperature, the return water temperature, the input power of the heat pump, the ambient temperature, and the ambient humidity.

[0015] The above solution analyzes the multi-dimensional data comprehensively, providing data support for the subsequent temperature prediction.

[0016] Optionally, the loss function is: ;

[0017] Among them, is the weight of the kth state data, is the predicted value of the kth state data, is the label of the kth state data, and K is the total number of state data.

[0018] The above setting of weights for each state data can correct the difference between the predicted value and the label, making the trained network prediction model more accurate.

[0019] Optionally, the coordination degree is the coupling relationship between any dimension time series and the target time series calculated by the coupling coordination degree model.

[0020] Optionally, the noise degree is:

[0021] ; where is the k-th data under the n-th dimension time series, is the mean value of the irregular changes corresponding to the n-th dimension time series, is the standard deviation of the irregular changes corresponding to the n-th dimension time series; the irregular changes are the fluctuations of the normal time series corresponding to the n-th dimension time series obtained by using the additive model or the multiplicative model.

[0022] The above solution can eliminate some normal fluctuations and obtain an accurate noise degree.

[0023] Optionally, the process of obtaining the noise degree is:

[0024] Centering on any data in any dimension time series, obtaining a time series segment under a set window, and calculating the standard deviation of all data in the corresponding time series segment; further obtaining the standard deviation of all data, and taking each standard deviation as the noise degree of the corresponding data.

[0025] Optionally, the comparison of the predicted temperature with the target outlet water temperature to achieve the control of the heat pump includes:

[0026] Obtaining the absolute value of the predicted difference between the predicted temperature and the target outlet water temperature;

[0027] When the absolute value of the predicted difference is less than the threshold, keep the operating state of the heat pump unchanged; when the absolute value of the predicted difference is greater than or equal to the threshold, control the operating state of the heat pump according to the control strategy; the control strategy is: when the predicted temperature is less than the target outlet water temperature, taking the sum of the predicted temperature and the set amount as the adjusted temperature value; when the predicted temperature is greater than the target outlet water temperature, taking the difference between the predicted temperature and the set amount as the adjusted temperature value.

[0028] The above solution can effectively adjust the temperature of the heat pump.

[0029] Optionally, the prediction network model is an LSTM model or a TCN model.

[0030] Optionally, the specific process of training the network prediction model is:

[0031] Input the training set into the network prediction model for training, calculate the loss value using the loss function, and adjust the parameters of the network prediction model using the gradient descent algorithm until the loss value between the predicted value and the label output is less than the threshold or the number of training times reaches the set number, then stop training and obtain the trained network prediction model.

[0032] In a second aspect, a heat pump control system based on state data provided by the present invention includes:

[0033] A processor;

[0034] A memory storing computer instructions for heat pump control based on state data, and when the computer instructions are run by the processor, the system executes the above-mentioned heat pump control method based on state data.

[0035] The beneficial effect of the present invention is that the solution of the present invention can make the loss during the training process closer to the real situation by improving the loss function in the network prediction model, improve the accuracy of the trained network prediction model, and further obtain an accurate predicted temperature to achieve effective control of the heat pump. Description of the Drawings

[0036] Figure 1 Schematically shows a flowchart of the steps of a heat pump control method based on state data in this embodiment;

[0037] Figure 2 Schematically shows a structural block diagram of a heat pump control system based on state data in this embodiment. Detailed Embodiments

[0038] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. The described embodiments are some of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative efforts fall within the protection scope of the present invention.

[0039] Taking the heat pump in the process of household heating or commercial building heating as an example, a heat pump control method based on state data provided in this embodiment will be introduced.

[0040] Specifically, as Figure 1 shown, a heat pump control method based on state data provided in this embodiment includes the following steps:

[0041] Step S1, collect the state data of the heat pump at the current moment.

[0042] The state data collected at the current moment in this embodiment includes data in multiple dimensions; specifically, the data in multiple dimensions includes, but is not limited to, the inlet temperature, the outlet temperature, the return water temperature, the input power of the heat pump, the ambient temperature, and the ambient humidity.

[0043] Specifically, the above data can be collected by installing temperature sensors on the heat pump and collecting the inlet temperature, the outlet temperature, and the return water temperature according to the sampling interval; installing temperature sensors and humidity sensors at appropriate positions in the target area and collecting the ambient temperature and the ambient humidity of the target area according to the sampling interval. The input power can be calculated by directly measuring the current and voltage of the heat pump power supply. When measuring, the measuring instrument needs to be installed on the circuit to record the current and voltage values of the heat pump during operation.

[0044] The sampling interval mentioned above can be 1 min or 5 min.

[0045] In this embodiment, the target outlet temperature of the heat pump is also obtained. Usually, the target outlet temperature is set according to the ambient temperature in the target area; that is, when the ambient temperature is lower, the setting of the target outlet temperature should be higher to ensure the effective operation of the heat pump; generally, the target outlet temperature is set between 55°C and 60°C.

[0046] Step S2, input the state data at the current moment into the network prediction model and output the predicted temperature at the next moment.

[0047] The network prediction model in this embodiment adopts an LSTM model or a TCN model.

[0048] Taking the LSTM model as an example, the specific training process includes steps S21 - S22, specifically:

[0049] Step 21, obtain the training set.

[0050] The training set in this embodiment is obtained by obtaining the state data within the historical set time series.

[0051] Among them, there are data in multiple dimensions at different moments, and the data in multiple dimensions at the same moment constitutes the state data. The data in the same dimension at different moments constitutes the dimension time series.

[0052] In this embodiment, the outlet temperature at the next moment of the moment where each state data is located in the historical set time series is also obtained, and this outlet temperature is used as the label corresponding to the state data. Then, there are multiple labels corresponding to the multiple state data within the historical set time series.

[0053] Among them, in this embodiment, the data of some of the above dimensions can be used for the training of the subsequent network prediction model; alternatively, the data of all the above dimensions can be used for the training of the network prediction model to improve the accuracy of model training.

[0054] Step 22: Input the sample data into the recurrent neural network model for training, calculate the loss value using the loss function, and adjust the parameters of the network model using the gradient descent algorithm to adjust the parameters of the network model until the loss value between the predicted value and the label output is less than the threshold or the number of training times reaches the set number of times, and a trained temperature prediction model is obtained.

[0055] The above threshold can be 0 or a value close to 0. Exemplarily, the threshold can be 0.1.

[0056] In order to obtain a more accurate network prediction model, the loss function in this embodiment is the mean square error loss with weights, specifically:

[0057] ;

[0058] Among them, is the loss value, is the weight of the k-th state data, is the predicted value of the k-th state data, is the label of the k-th state data, and K is the total number of state data.

[0059] The process of obtaining the above weights is as follows:

[0060] First, calculate the coordination degree between any dimension time series and the target time series. Among them, the target time series is the time series composed of all labels.

[0061] Specifically, the coordination degree is the coupling relationship between any dimension time series and the target time series calculated through the coupling coordination degree model.

[0062] The coupling coordination degree model (CCDM) is a model used to measure the coupling degree and coordinated development level between internal elements of a system, and it can measure the overall operating state and the mutual relationship between elements. Therefore, in this embodiment, the coupling relationship between any dimension time series and the target time series is calculated to determine the strength of the correlation between the two.

[0063] Specifically, the coordination degree is the square root of the product of the pre-calculated coupling degree C value and the coordination index T value.

[0064] Since the calculation of the coupling degree C value and the coordination index T value in the coupling coordination degree model is prior art, it will not be elaborated here too much.

[0065] Secondly, obtain the noise level of each data in each dimension time series.

[0066] In one embodiment, the process of obtaining the noise level is as follows:

[0067] Use an additive model or a multiplicative model to obtain the irregular fluctuations of the normal time series corresponding to each dimension time series;

[0068] Obtain the mean and standard deviation of the irregular fluctuations;

[0069] According to the mean and standard deviation of the irregular fluctuations, calculate the noise level of each data in the corresponding dimension time series.

[0070] Specifically, the noise level is expressed as:

[0071] ; where is the k-th data under the n-th dimension time series, is the mean of the irregular fluctuations corresponding to the n-th dimension time series, is the standard deviation of the irregular fluctuations corresponding to the n-th dimension time series.

[0072] Among them, the irregular fluctuations are the fluctuations of the normal time series corresponding to the n-th dimension time series obtained by using an additive model or a multiplicative model. Since the above additive model or multiplicative model is a prior art, the specific process of the irregular fluctuations will not be elaborated here too much.

[0073] It should be noted that the normal time series is a time series of known non-abnormal data, and this normal time series can be obtained from the data of each dimension collected in the historical operation process of the heat pump. Each dimension time series corresponds to a normal time series of the corresponding dimension. Therefore, each dimension time series corresponds to an irregular fluctuation.

[0074] The reason for obtaining the irregular fluctuations of the normal time series is that considering that there may be some small noises in the data of each dimension during collection due to the influence of the environment, the sensor itself, etc., and this data is not real noise data or abnormal data. Therefore, it is necessary to obtain the normal fluctuation component in the normal time series to determine whether the data of each dimension is within the normal range.

[0075] In another embodiment, the process of obtaining the noise level can also be: taking any data in any dimension time series as the center, obtaining the time series segment under the set window, and calculating the standard deviation of all data in the corresponding time series segment; then obtaining the standard deviation of all data, and taking each standard deviation as the noise level of the corresponding data. The set window can be set according to the actual situation. Exemplarily, the set window is 11 or 15.

[0076] The above noise degree is used to evaluate the possibility that the data in each dimension may be noise. If the corresponding noise degree of the data is larger, the possibility that the data belongs to noise is greater, and its importance in the subsequent network prediction model should be smaller, so as to ensure that the trained network prediction model is more accurate.

[0077] Then, based on the coordination degree and the noise degree, the weights are obtained.

[0078] Among them, the weights are: ; is the weight of the k-th state data, is the coordination degree of the data in the n-th dimension of the k-th state data, is the noise degree of the data in the n-th dimension of the k-th state data, N is the total number of dimensions in the state data, and exp( ) is the exponential function with the natural constant e as the base.

[0079] Among them, when the coordination degree is smaller, it proves that the correlation between the time series of the n-th dimension and the target time series is weaker, that is, the influence of each data in the time series of the n-th dimension on the corresponding data in the target time series is smaller, then the weight is smaller; when the coordination degree is 0, at this time the time series of the n-th dimension and the target time series are completely independent and have no correlation, and at this time the weight is smaller. When the noise degree is smaller, it proves that the possibility that the data in the n-th dimension of the k-th state data belongs to noise is smaller, and at this time the data in this dimension is more important, and the weight is larger.

[0080] The purpose of obtaining the above weights is that when collecting state data, there are different strong and weak correlation situations between the data in each dimension and the target time series (labels), and the data collected in different dimensions may have different noise situations due to the influence of the environment and the sensors themselves. Therefore, it is necessary to consider both the importance of the data in different dimensions to the target time series and whether there is noise in the data in each dimension, so as to use the weights to correct the difference change between the predicted value and the label of the data in the corresponding dimension, so that the loss value in the training process is more accurate, and then a more accurate network prediction model is obtained.

[0081] It should be noted that before training, it is also necessary to preprocess the state data within the historical set time series, such as data cleaning, data denoising, etc.

[0082] In this embodiment, after obtaining the trained network prediction model, the state data at the current moment is input into the network prediction model, and the predicted temperature at the next moment of the current moment can be obtained, realizing the prediction of the outlet temperature of the heat pump.

[0083] In step S3, the predicted temperature is compared with the target outlet water temperature to control the heat pump. Specifically, based on obtaining the absolute value of the predicted difference between the predicted temperature and the target outlet water temperature, the operating state of the heat pump is determined. That is, when the absolute value of the predicted difference is less than the threshold, the operating state of the heat pump remains unchanged. The threshold can be set to 1°C, and of course, it can also be set according to the actual situation.

[0084] When the absolute value of the predicted difference is greater than or equal to the threshold, the operating state of the heat pump needs to be controlled. Specifically, there are two cases in the control process. The first case is that when the predicted temperature is less than the target outlet water temperature, it is adjusted upward by a set amount based on the predicted temperature to obtain an adjusted temperature value. When the predicted temperature is greater than the target outlet water temperature, it is adjusted downward by a set amount based on the predicted temperature data to obtain an adjusted temperature value.

[0085] The set amount can be the absolute value of the predicted difference. Of course, it can also be reset. For example, the set amount can be 1°C or 2°C.

[0086] The above control strategy can precisely control the operating state of the heat pump. It should be noted that the adjusted temperature value in this embodiment is used to pre-adjust the temperature at the next moment of the current moment.

[0087] When controlling and adjusting the operating state of the heat pump in the above embodiment, corresponding staff can also be reminded to perform the control and adjustment based on the staff's experience.

[0088] Furthermore, for the case where the absolute value of the predicted difference is greater than or equal to the threshold, there may also be a situation where the prediction effect of the trained network prediction model is not good. At this time, the network prediction model needs to be evaluated. Specifically, by obtaining the accuracy rate of the network model to evaluate the network prediction model. That is, when the accuracy rate is greater than the accuracy rate threshold, the network prediction model is accurate. When the accuracy rate is less than or equal to the accuracy rate threshold, relevant staff can be notified in a timely manner to control the heat pump based on the staff's experience.

[0089] Among them, the accuracy rate threshold is 0.95, and of course, it can also be set according to the actual situation.

[0090] In the present invention, the meaning of "a plurality of" is at least two, such as two, three or more, unless otherwise specifically defined.

[0091] The solution of the present invention can train a network prediction model based on the collected state data of the heat pump and optimize the loss function of the network prediction model to obtain a more accurate network prediction model, and further obtain an accurate predicted temperature. That is, the solution of the present invention can optimize the operating state of the heat pump and improve the self - adaptability and intelligent level of the heat pump.

[0092] The present invention also provides a heat pump control system based on state data. As Figure 2 shown, the control system includes a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the heat pump control method based on state data according to the present invention is implemented.

[0093] The control system further includes a communication bus, a communication interface and other components well - known to those skilled in the art. Their settings and functions are known in the art, so they will not be described in detail here.

[0094] In the present invention, the aforementioned memory can be any tangible medium that contains or stores a program, and this program can be used by or in combination with an instruction execution system, device or device. For example, a computer - readable storage medium can be any suitable magnetic storage medium or magneto - optical storage medium. For example, resistive random - access memory (RRAM), dynamic random - access memory (DRAM), static random - access memory (SRAM), enhanced dynamic random - access memory (EDRAM), high - bandwidth memory (HBM), hybrid memory cube (HMC), etc., or any other medium that can be used to store the required information and can be accessed by an application, a module or both. Any such computer storage medium can be part of the device or accessible or connectable to the device. Any application or module described in the present invention can be implemented by computer - readable / executable instructions stored or otherwise held by such a computer - readable medium.

[0095] Although the specification has described multiple embodiments of the present invention, such embodiments are provided only by way of example. In practice, various alternative solutions to the embodiments of the present invention described herein can be adopted.

Claims

1. A heat pump control method based on state data, characterized in that: include: Get the training set; The training set is a plurality of state data and labels of each state data in a historical set time series, wherein the label is the water outlet temperature at the next moment of the moment of each state data; the state data includes data of multiple dimensions; Using the training set to train a network prediction model; The loss function in the training process is the mean square error loss with weights; the weights for: ; is the coordination degree of the data of the nth dimension in the kth state data, is the noise degree of the data of the nth dimension in the kth state data, N is the total number of dimensions in the state data, and exp() is an exponential function; the coordination degree represents the coupling relationship between the time series of any dimension and the target time series, the target time series is the time series composed of all tags, and the time series of any dimension is the time series composed of data of the same dimension; the noise degree represents the situation that the data of each dimension belongs to noise; The state data of the heat pump at the current moment is input into the network prediction model to obtain the predicted temperature at the next moment; the predicted temperature is compared with the target water outlet temperature to realize the control of the heat pump.

2. A heat pump control method based on state data according to claim 1, characterized in that: The data of the multiple dimensions include but are not limited to water inlet temperature, water outlet temperature, return water temperature, input power of the heat pump, ambient temperature and ambient humidity.

3. A heat pump control method based on state data according to claim 2, characterized in that: The loss function for: ; in, is the weight of the k-th state data, is the predicted value of the kth state data, is the label of the kth state data, and K is the total number of state data.

4. A heat pump control method based on state data according to claim 1, characterized in that: The coordination degree is the coupling relationship between any dimensional time series and the target time series calculated by the coupling coordination degree model.

5. A heat pump control method based on state data according to claim 1, characterized in that: The noise level for: ;in, is the kth data in the nth dimension time series, is the mean of the irregular changes corresponding to the n-th dimension time series, is the standard deviation of the irregular variation corresponding to the n-th dimensional time series; the irregular variation is the fluctuation of the normal time series corresponding to the n-th dimensional time series obtained by using an additive model or a multiplicative model.

6. A heat pump control method based on state data according to claim 1, characterized in that: The process of obtaining the noise level is as follows: Taking any data in any dimensional time series as the center, obtain the time series segment under the set window, and calculate the standard deviation of all data in the corresponding time series segment; then obtain the standard deviation of all data, and use each standard deviation as the noise degree of the corresponding data.

7. A heat pump control method based on state data according to claim 1, characterized in that: The predicted temperature is compared with the target outlet water temperature to control the heat pump, including: Obtain the absolute value of the predicted difference between the predicted temperature and the target water outlet temperature; When the absolute value of the predicted difference is less than a threshold value, the operating state of the heat pump is kept unchanged; when the absolute value of the predicted difference is greater than or equal to the threshold value, the operating state of the heat pump is controlled according to the control strategy; the control strategy is: when the predicted temperature is less than the target water outlet temperature, the sum of the predicted temperature and the set amount is used as the adjusted temperature value; when the predicted temperature is greater than the target water outlet temperature, the difference between the predicted temperature and the set amount is used as the adjusted temperature value.

8. The heat pump control method based on state data according to claim 1, characterized in that: The network prediction model is an LSTM model or a TCN model.

9. A heat pump control method based on state data according to claim 8, characterized in that: The specific process of using the training set to train the network prediction model is as follows: The training set is input into the network prediction model for training, and the loss function is used to calculate the loss value. The parameters of the network prediction model are adjusted using the gradient descent algorithm until the loss value between the output prediction value and the label is less than the threshold or the number of training times reaches the set number. The training is stopped and a trained network prediction model is obtained.

10. A heat pump control system based on state data, characterized in that: include: processor; A memory storing computer instructions for controlling a heat pump based on state data. When the computer instructions are executed by the processor, the system executes a method for controlling a heat pump based on state data according to any one of claims 1 to 9.

Images