Method for data-driven predictive control of a heat pump system, computing unit and heat pump system

Abstract

The invention relates to a method for data-driven predictive control of a heat pump system, in which at least one future input trajectory and in particular at least one future output trajectory of the heat pump system are ascertained at least on the basis of system data of the heat pump system measured at an earlier time using a defined objective function, the system data comprises at least input measurement data and output measurement data of the heat pump system, a control equation filled with the system data is solved by means of the computing unit, and a trajectory of the heat pump system is determined by means of the control equation according to a matrix-vector multiplication of at least one measurement data matrix and a decision variable vector, the measurement data matrix is composed of at least two Hankel matrixes which are stacked up and down and respectively comprise one measurement data type of system data. In order to take into account the operating lower limit of the heat pump system, binary variables are assigned to at least a plurality of elements of a future input trajectory in the control equation, respectively.

Description

Technical Field

[0001] The invention relates to a method for data-driven predictive regulation of a heat pump system, a computing unit and a heat pump system. Background Art

[0002] A method for data-driven predictive regulation of energy systems is known from the scientific publication “Towards data-driven predictive regulation of multi-energy distribution systems” by D. Bilgic et al. (Electric Power Systems Research, 212(2): 108311). Summary of the Invention

[0003] The present invention proceeds from a method for data-driven predictive control of a heat pump system, wherein, using a defined target function, in particular a defined target function to be minimized, at least one future input trajectory of the heat pump system and in particular a future output trajectory of the heat pump system are determined based on system data of the heat pump system measured at an earlier time, wherein the system data include at least input measurement data and output measurement data of the heat pump system, wherein, for data-driven predictive control of the heat pump system, a control equation filled with the system data is solved by means of a calculation unit of the heat pump system or by means of a calculation unit arranged outside the heat pump system and in communication with the heat pump system, wherein the calculation unit solves a control equation filled with the system data by means of A control equation is provided for determining a trajectory of the heat pump system, in particular at least an input trajectory and at least one output trajectory of the heat pump system, based on a matrix-vector multiplication of at least one measurement data matrix and a decision variable vector, wherein the measurement data matrix is ​​composed of at least two Hankel matrices stacked one above the other, each Hankel matrix including only one measurement data type of the system data (e.g., only one input measurement data type, such as electrical power or forward flow temperature of the water side (system to be heated); or only one output measurement data type, such as generated heat flow, which can be calculated, for example, from the measured temperature difference and mass flow on the condenser water side (system to be heated), or from the measured temperature difference and mass flow on the refrigerant side of the condenser (heat pump). Alternatively, the regulated flow temperature and mass flow (both on the condenser water side) can also be selected as the output measurement data. In this case, the measurement data matrix will include at least three Hankel matrices stacked one above the other, wherein the two Hankel matrices will each include a different output measurement data type of the system data.

[0004] It is proposed that, in order to take into account the lower operating limit (Betriebsschranke) of the heat pump system, in particular a non-zero lower operating limit, at least a plurality of elements (Eintrag) of the future input trajectory in the control equation and preferably also the initial input value of the input trajectory are each assigned a binary variable, in particular by multiplication, wherein the elements of the future input trajectory respectively correspond in particular to the prediction time step of the predicted input of the heat pump system. This can advantageously improve the data-driven control of the heat pump system. Advantageously, the control can be limited to an operating range of the heat pump system that can be operated technically and / or economically reasonably. Advantageously, the control can be limited only to the following operating range of the heat pump system: the operating range includes the normal operating range of the heat pump of the heat pump system between the minimum partial load and the full load and the shutdown state. Advantageously, since in particular the function and / or operation can be optimized and / or the efficiency can be improved, costs and / or energy consumption can be reduced. Advantageously, since in particular the computing power required for the control can be reduced, costs and / or energy can be saved. Advantageously, the use of the proposed binary variables allows the regulation of the heat pump to be optimized, since when determining the future input trajectory of the heat pump system, in particular when regulating the heat pump system, a rapid switch between normal operation above a minimum power and shutdown (in particular at least in no-load operation) can be taken into account.

[0005] In particular, in data-driven predictive control of system devices, physical modeling of the system devices is replaced by predicting the system behavior using system data (in particular, at least input measurement data and output measurement data). Thus, in data-driven predictive control, physical modeling of the actual system devices can be advantageously bypassed, and system behavior can be predicted solely using measurement data. A heat pump system is particularly configured to absorb thermal energy from a relatively low-temperature energy storage device (e.g., ambient air, ground, etc.) while consuming technical work, and transfer this energy, along with drive energy (in particular, given by an input trajectory), as useful heat to a system to be heated (e.g., a building) at a higher temperature. To this end, the heat pump system particularly includes at least one heat pump, which may include various components, such as a compressor, a condenser, a throttle valve, and / or an evaporator. Furthermore, the heat pump system may also include a computing unit and various heat exchangers, among others. Preferably, the input trajectory determined by the data-driven predictive control includes the electrical power supplied to the heat pump system or a power request (power demand) for the heat pump system. Preferably, in the data-driven predictive regulation, the power request of at least one heat pump of the heat pump system is regulated, in particular in order to obtain a certain heat flow at a given temperature on the evaporator side and the condenser side of the heat pump. In particular, in the proposed data-driven predictive regulation, subordinate regulations of the power request regulation (in particular the refrigeration cycle regulation of the heat pump) are not taken into account and preferably not performed. In particular, in the data-driven predictive regulation, in addition to the input trajectory, an output trajectory is also determined. It is also conceivable to determine further trajectories in the data-driven predictive regulation, such as further input trajectories or further output trajectories. The output trajectory in particular includes a heat flow output by the heat pump system, which is preferably determined based on a measured temperature difference and a mass flow on the water side of the condenser of the heat pump system, for example. The trajectory preferably includes a plurality of elements of time steps that follow one another in time, preferably regularly. The time steps of the input trajectory correspond to the time steps of the output trajectory.

[0006] In particular, the input trajectory and output trajectory are obtained at least to a large extent based on a system representation of the heat pump system, in particular a control equation representation of the heat pump system, which is based on system data measured at an earlier time of the heat pump system. The system data of the heat pump system, in particular the system representation (preferably the control equation representation) at least includes a value pair of output measurement data and input measurement data of the heat pump system. The input measurement data can be constructed as the electrical power input (operating load value / input load value) of the heat pump system. For example, the input measurement data can be constructed as the electrical power (or power demand) flowing into the compressor of the heat pump system and / or the forward flow temperature of the water side of the condenser of the heat pump system. The output measurement data is particularly constructed as heat flow measurement data corresponding to the input measurement data, which is preferably obtained indirectly, for example, on the condenser side of the heat pump system. Instead of value pairs, it is conceivable to incorporate other data (for example, exogenous measurement data) to form value triplets, value quadruplets, etc. These value clusters (Wertebündel) preferably constitute the system data. In particular, the system data is obtained at a time before regulation, for example, when manufacturing, calibrating or installing the heat pump system. It is conceivable that the heat pump system (particularly during operation) can update the system data autonomously, at least partially at certain regular or irregular times. For this purpose, the heat pump system preferably has corresponding sensors. It is also conceivable that the heat pump system can (particularly autonomously or under the guidance of an operator) perform a plausibility check on the system data, thereby being able to detect, in particular, system changes that may require an update of the system data.

[0007] Preferably, in data-driven predictive regulation, at least one future input trajectory and preferably also a future output trajectory is determined by minimizing an objective function (in particular a loss function, "cost function" in English). The objective function reflects, for example, technical and / or economic relationships. The objective function can minimize the electricity costs related to the time of day and / or achieve the heat demand related to the time of day. For example, the objective function can be optimized by minimizing the energy costs of the operation while maintaining a minimum temperature (which may be related to the time of day). In particular, the goal of the optimization is to design the operation of the heat pump system so that, for example, energy costs and / or energy consumption are minimized. As a result, for example, the electricity demand during load peaks of national electricity demand can be reduced, so that the heat pump system can advantageously contribute to improving grid stability. Alternatively or additionally, efficiency can also be improved and / or the desired temperature level can be better maintained. In data-driven predictive regulation, the objective function is minimized while taking into account the upper and lower operating limits defined for the heat pump system.

[0008] In particular, the lower operating limit of the heat pump system corresponds to the minimum partial load of the heat pumps of the heat pump system, which preferably corresponds to a power demand greater than zero. For example, the power demand of the heat pump at the lower operating limit may correspond to approximately 15%, approximately 20%, or approximately 25% of the power demand at the full load (upper operating limit) of the heat pump. To account for the lower operating limit of the heat pump system, the data-driven predictive control of the heat pump system, in particular the data-driven predictive control, is limited to a first operating load range of the heat pumps of the heat pump system and a second operating load range of the heat pumps of the heat pump system, wherein the first operating load range and the second operating load range do not overlap and are spaced apart from each other. In particular, the operating load ranges to which the data-driven predictive control is limited are each configured as a range of input power values ​​(power requests) that can be input by the heat pumps of the heat pump system, each of which is bounded by an upper limit and a lower limit. However, it is also conceivable that at least one operating load range is defined by only one operating point, i.e., in particular, for one operating load range, the upper limit may be the same as the lower limit. In particular, the upper limit of the first operating load range is lower than the lower limit of the second operating load range. Preferably, the first operating load range only includes operating loads / power demands of zero. In particular, when the operating load is set to zero, the heat pump is in an idling state. In particular, when the operating load is set to zero, the heat pump is deactivated or enters a standby state. In particular, the first operating load range does not include any operating load values / input power values ​​that are also within the second operating load range. In particular, the operating load ranges constitute a continuous operating load range. Preferably, the continuous operating load range does not include excluded operating load values / input load values. Preferably, the first operating load range and the second operating load range are respectively adjacent to an exclusion range, which is located between the first operating load range and the second operating load range. Preferably, the upper limit of the first operating load range and the lower limit of the second operating load range delimit the exclusion range. The exclusion range, in particular, constitutes a range of operating load values / input load values ​​that are not allowed / cannot be set by regulation. In particular, at operating load values / input load values ​​in the exclusion range, unstable (oscillating) operation of the heat pump of the heat pump system is expected. “Provided” is to be understood as specifically programmed, designed and / or equipped. Providing an object for a specific function is to be understood in particular to mean that the object realizes and / or carries out the specific function in at least one application state and / or operating state.

[0009] Advantageously, expressing the heat pump system's control equations allows for efficient, particularly analytically solvable, regulation of the heat pump system. Preferably, the computing unit includes at least a solver for solving the control equations. Alternatively, an external computing unit, particularly a central computing unit (e.g., a cloud, etc.), may include and / or have access to at least a solver for solving the control equations. Preferably, the solver solves the entire optimization problem: that is, it minimizes the objective function while taking into account the control equations (matrix-vector multiplication with the trajectory on the right side and binary variables) and the constraints / operating limits of the heat pump system. The control equations may also form or be expressed as a system of control equations. In particular, the control equations include at least one measurement data matrix and a decision variable vector on one side (left side) of the equation and a trajectory vector on the other side (right side). Preferably, the control equations can be expressed as matrix equations. A "computing unit" is to be understood, in particular, as a unit comprising at least one control electronics unit. "Control electronics" is to be understood, in particular, as a unit comprising a processor and an electronically readable memory, as well as an operating program stored in the memory. A “solver” is to be understood as meaning, in particular, a special mathematical computer program which can solve mathematical equations and / or systems of equations and / or terms and / or similar tasks.

[0010] In the data-driven predictive control of the heat pump system, at least a plurality of time steps of all (non-fixedly predetermined) trajectories of the input trajectory, in particular the trajectory vector, are preferably optimized simultaneously. In particular, in the method for controlling the heat pump system, the trajectory vector can be repeatedly determined and / or updated every fixed time step or after each time step. In particular, at least the future input trajectory determined in the control constitutes a prediction of the heat pump system within the prediction range. The control equations represent possible system operation plans for future regulation within a fixed timeframe, with the prediction horizon encompassing a fixed number of time steps. In particular, the control equations constitute equations, in particular systems of equations, for a data-driven system representation of a heat pump or heat pump system. In particular, multiple measurement data types can be incorporated into the measurement data matrix for inputs, outputs, and (if necessary) exogenous inputs. Accordingly, the trajectory vectors on the right-hand side of the control equations can also contain corresponding trajectories for these multiple measurement data types.

[0011] The known data-driven predictive system description already includes the regulation equation (1) with a regulation matrix consisting of a plurality of vertically stacked Hankel matrices with known matrix notation. These Hankel matrices of the known regulation matrix can include, on the one hand, the input measurement data u d and on the other hand includes outputting the measurement data y dSpecifically, the measurement data matrix includes a separate Hankel matrix for each measurement data type, which is arranged vertically within the measurement data matrix. In the known regulation equation, these Hankel matrices are interconnected via the decision variable vector g to determine a trajectory vector containing the input trajectory u and the output trajectory y. The input trajectory u and the output trajectory y each constitute a vector.

[0012]

[0013] Known control equations do not allow to exclude operating load ranges from the control. The control equation according to the invention is expressed as equation (2) below. In order to perform data-driven predictive control of the heat pump system, the solver preferably solves the following equation (2).

[0014]

[0015] The left side of equation (2) is a measurement data matrix and a decision variable vector, the measurement data matrix having two Hankel matrices stacked one above the other. In the case of additional consideration of exogenous measurement data, for example, another Hankel matrix containing exogenous measurement data can be inserted between the Hankel matrices shown in equation (2). The right side of equation (2) is a trajectory vector having input trajectories and output trajectories. In the case of consideration of exogenous measurement data, the trajectory vector also includes an additional exogenous data vector. The above Hankel matrix contains the input measurement data The Hankel matrix below contains the output measurement data In formula (2), no additional exogenous measurement data are taken into account. In this case, the superscript d indicates that this relates to input measurement data of the heat pump measured at an earlier moment. It is conceivable that the (total) measurement data matrix has a plurality of Hankel matrices with temporally separated (shorter) trajectories. These Hankel matrices (with the same data type, but with trajectories from different past time intervals) are then stacked horizontally on top of each other. In this case, the subscript i indicates to which relevant measurement data group the individual measurement data elements belong that are calculated within a common time step. The subscript i extends from the value 1 (the first time step of the first measurement data group / trajectory) to the value T (the last time step of the trajectory). N indicates the prediction horizon that should be adjusted. The prediction horizon N indicates, in particular, how far into the future the situation should be considered in each iteration of the data-driven predictive adjustment. The prediction horizon indicates the number of future input trajectory values ​​that should be calculated in each iteration of the adjustment. For example, in the case of hourly readjustment of the heat pump, the prediction horizon can take the value N=12 or N=24. In this case, T ini Indicates the number of known initial values ​​that should be included in the regulation. In particular, at least the initial values, for example the corresponding current (measured) values, are respectively and y ini Incorporate into the regulation process. This case is described exemplarily in formula (2). Alternatively, it is also conceivable to incorporate other measured values ​​from the past immediately before the initial value into the regulation. In this case, there is T ini >1. In this case, T corresponds to the length of the corresponding measurement data trajectory. The measurement data trajectory has, in particular, T elements. In particular, the vertical arrangement of the Hankel matrix within the measurement data matrix corresponds to the vertical arrangement of the trajectory vectors. The number of elements of the decision variable vector corresponds, in particular, to the number of columns of the measurement data matrix. The number of elements of the trajectory vector corresponds, in particular, to the number of rows of the measurement data matrix.

[0016] J.Coulson et al. published a scientific publication entitled "Data-Enabled Predictive Control: In the Shallows of the DeePC" (arXiv:1811.05890) and proposed that the so-called slack variable σ y Expanding the regulation equation (2), in this case, the slack variable is added to the initial output (element y of the output trajectory y ini ). Obviously, with slack variables σ y The regulation equation of should also be regarded as the regulation equation according to the present invention. The slack variable σ can also be optimized when solving the regulation equation. y . It is also possible to consider, and especially optimize, the slack variable σ when minimizing the objective function y .

[0017] The decision variable vector contains elements g l , where index 1 corresponds to the column of the measurement data matrix. Therefore, the index l specifically represents the column number of the measurement data matrix. The trajectory vector contains the elements In this case, the superscript x indicates the trajectory to which the element of the trajectory vector belongs. In this case, the subscript k corresponds to the individual time steps of the input trajectory in the prediction horizon / future. The input trajectory is denoted by u and the output trajectory is denoted by y. The subscript k is a consecutive number from 1 to N. Here, the elements It is also possible to enter the trajectory u k δ k The indices ini and 1 to N represent the initial time step "ini" (whose measured values ​​have already been determined) and the future time steps 1 to N to be determined during the control. The decision variable vector g can also be optimized when solving the control equations. The decision variable vector g can also be considered, and in particular optimized, when minimizing the objective function.

[0018] Binary variable δ kcan only take the value 0 or 1 (see also exemplary equation (3)). By modifying the elements of the input trajectory with a binary variable, the corresponding element is either set to zero (δ k =0), or remain unchanged (δ k =1).

[0019]

[0020] In the control equation, preferably all elements of the future input trajectory are each (individually) assigned a binary variable, in particular by multiplication. In particular, with the exception of the elements of the future input trajectory, in particular the initial values ​​of the future input trajectory and the preferred input trajectory (which describe the input power, preferably the power demand of the heat pump system), no other elements of the control equation are assigned a binary variable. It is conceivable that the control equation may include other input trajectories that describe input parameters other than the input power / power demand of the heat pump system. Preferably, such other input trajectories are not assigned a binary variable. By using binary variables in accordance with the present invention, it is advantageous to combine known data-driven predictive control with the advantages of mixed-integer decision making. This advantageously expands the known control method by allowing the heat pump to be operated only within a normal operating range between partial load and full load or to be shut down. This means, in particular, that in the data-driven predictive control according to the present invention, it is possible to switch back and forth between the normal operating range of the heat pump (for which upper and lower operating limits are predefined) and completely shutting down the heat pump, depending on the predicted heat demand. Preferably, the output dynamics (eg, the delay in the thermal output) when the heat pump is switched off are reflected in the system description of the proposed equations, in particular the regulation equation (2).

[0021] Furthermore, it is proposed that when solving the control equations by means of a computing unit, in particular a solver, in particular when minimizing the objective function, at least the elements of the future input trajectory to which binary variables are assigned are optimized, preferably at least the elements of the future input trajectory to which binary variables are assigned are optimized together with the corresponding binary variables. This can advantageously improve the functionality of the heat pump system and / or save costs.

[0022] Additionally, it is proposed that when solving the control equations by means of a computing unit, in particular a solver, in particular when minimizing a target function, at least elements of the future output trajectory corresponding to time steps of the future input trajectory to which a binary variable is respectively assigned are optimized, and / or when solving the control equations by means of a computing unit, in particular a solver, in particular when minimizing the target function, at least decision variables of a decision variable vector are optimized. This can advantageously improve the functionality of the heat pump system and / or save costs.

[0023] Furthermore, it is proposed that, for solving the control equations, in particular minimizing the target function, by means of a computing unit, in particular a solver, the elements of the measurement data matrix remain unchanged and preferably are not optimized. This advantageously allows for an efficient solution. As already mentioned, the elements of the measurement data matrix can be changed, in particular recalibrated, between the time-spaced solutions of the control equations.

[0024] Furthermore, it is proposed that, in order to solve the control equation, in particular to minimize the objective function, by means of a calculation unit, in particular by means of a solver, the elements of the trajectory modified by binary variables, in particular the elements of the future input trajectory modified by binary variables, are replaced by two inequalities. This can advantageously reduce the computational workload, because the solution of the control equation can be simplified to a "mixed integer linear programming problem" in particular. By introducing the inequalities, the problem to be solved, in particular for the solver, is advantageously relaxed. By introducing the inequalities, the problem to be solved by the solver is advantageously linearized. This can advantageously improve and / or accelerate the control process. Preferably, each element of the input trajectory on the right side of the control equation is replaced by two inequalities. In particular, in order to solve the control equation, in particular to minimize the objective function, it is preferred that, in order to relax the control equation, in particular by the calculation unit, the elements of the input trajectory modified by binary variables are replaced by parameters determined by two inequalities. The trajectory elements of the inequality are in particular determined by the parameters , which is determined by the inequality given in the following formula (4).

[0025]

[0026] The following effect will occur in particular according to the expression of formula (4): when the binary variable δ k = 0, the trajectory element The value of is restricted to 0 both upward and downward; when the binary variable δ k =1, the trajectory element The value of is always higher than the lower operating limit of the heat pump system. In addition, formula (4) preferably produces the following effect: when the binary variable δ k =1, the trajectory element The value of is determined / limited by two operating limits (the lower limit and the upper limit of the second operating load range). On the one hand, the introduction of a binary variable advantageously allows for rapid switching of the heat pump operation between the normal operating range above the minimum power of the heat pump and the complete shutdown of the heat pump. In addition, since the proposed control equation can take into account the dynamics of the heat pump shutdown (e.g. caused by the delay of the heat pump power), this leads to advantageous cost savings in the operation of the heat pump system. On the other hand, the introduction of inequality (3) advantageously converts the optimization problem into a linear representation, which can make the solver solution more efficient in particular because the mathematical problem to be solved becomes less complex. This means that, advantageously, either less computing power is required when solving, or more accurate regulation can be achieved when the computing power remains unchanged. This is particularly due to: for example, a larger prediction range can be considered in the linearized problem, and this can correspond to the same computing power requirements as a nonlinear problem with a smaller prediction range. The larger the calculated prediction range (at the same step length), the more accurate the regulation obtained thereby.

[0027] In this context, it is also proposed that the two inequalities be designed such that, when the binary variable is 1, the permissible values ​​of the trajectory, in particular the associated elements of the future input trajectory, for solving the control equations by means of the computing unit are limited by two respectively different and non-zero limit values, in particular an upper limit and a lower limit. This advantageously allows the creation of a linearized optimization problem, in which operational constraints can also be taken into account.

[0028] If these limit values ​​correspond to the permissible (input) operating range of the heat pump system (part load to full load), the heat pump-specific operating limits of the relevant heat pump of the heat pump system can advantageously be taken into account as operating limits.

[0029] Furthermore, if the two inequalities are designed such that, when the binary variable is 0, the permissible values ​​of the trajectory, in particular the associated elements of the future input trajectory, for solving the control equations by means of the computing unit are set to 0, then low-load operating ranges of the heat pump of the heat pump system, in which the heat pump cannot be operated or cannot be operated reliably / operationally reliably, can be advantageously excluded in the data-driven predictive control, while at the same time the possibility of completely shutting down the heat pump is still taken into account in the data-driven predictive control. This can significantly increase the efficiency of the control.

[0030] The aforementioned advantages are particularly effective if the value 0 corresponds to a deactivation of at least one heat pump of the heat pump system, preferably all heat pumps of the heat pump system, preferably the (entire) heat pump system, in particular at the respectively relevant prediction time step.

[0031] Furthermore, a computing unit and / or a heat pump system are proposed, which are configured for data-driven predictive regulation of a heat pump system having at least one heat pump and at least one computing unit. This advantageously improves the data-driven regulation of the heat pump system. Advantageously, costs and / or energy consumption can be reduced, as, in particular, functionality and / or operation can be optimized and / or efficiency can be increased. Advantageously, costs and / or energy can be saved, as, in particular, the computing power required for regulation can be reduced.

[0032] The method according to the invention, the computing unit according to the invention, and the heat pump system according to the invention are not limited to the above-described applications and embodiments. To achieve the functionalities described herein, the method according to the invention, the computing unit according to the invention, and the heat pump system according to the invention may, in particular, have a number of individual elements, components, units, and method steps that differs from the number mentioned here. Furthermore, for value ranges specified in this disclosure, the values ​​within the stated limits are also to be considered disclosed and can be used in any manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Further advantages can be derived from the following description of the drawings. The drawings illustrate an embodiment of the present invention. The drawings, the description, and the claims contain numerous combinations of features. Those skilled in the art will appropriately consider these features individually and combine them into other meaningful combinations.

[0034] Figure 1 A schematic diagram of a heat pump system is shown having an apparatus for data-driven predictive regulation of the heat pump system;

[0035] Figure 2 A schematic flow chart illustrating a method for data-driven predictive regulation of a heat pump system;

[0036] Figure 3a shows a schematic input power-time graph of a heat pump system;

[0037] Figure 3b A schematic output power-time graph of a heat pump system is shown;

[0038] Figure 4a Schematic input measurement data versus time graph showing system data used for data-driven predictive regulation of a heat pump system;

[0039] Figure 4b a schematic exogenous measurement data versus time graph illustrating system data used for data-driven predictive regulation of a heat pump system; and

[0040] Figure 4cSchematic output measurement data versus time graph showing system data used for data-driven predictive regulation of a heat pump system. DETAILED DESCRIPTION

[0041] Figure 1 A schematic diagram of a heat pump system 10 is shown. The heat pump system 10 has a device for data-driven predictive regulation of the heat pump system 10. The device is configured to perform a method for data-driven predictive regulation of the heat pump system 10. The heat pump system 10 has an integrated computing unit 22. Alternatively or additionally, the heat pump system 10 may also have a computing unit 22' arranged outside the heat pump system 10, which has a communication connection 70 with the heat pump system 10. The heat pump system 10 includes a heat pump 36. The heat pump 36 has a compressor, a condenser, a throttle valve and / or an evaporator and other components required for the operation of the heat pump 36. The heat pump 36 is configured as a closed refrigeration cycle. The heat pump 36 has a heat exchanger (not shown). In this example, the heat exchanger is a condenser. The heat exchanger can be configured to transfer the heat generated in the closed refrigeration cycle to heated water (e.g., a floor heating device). However, other applications of the heat pump 36 are also conceivable. The device is configured to determine a future input trajectory 28 (see FIG3 ) of the heat pump system 10 based on system data 30 measured at an earlier time of the heat pump system 10 using a defined objective function to be minimized. An exemplary objective function is given in the following formula (5):

[0042]

[0043] Figure 2 A schematic flow chart of a method for data-driven predictive regulation of a heat pump system 10 is shown. A computing unit 22, 22' is provided for carrying out the method. In at least one method step 24 (which is not necessarily part of the regulation method), system data 30 of the heat pump system 10 are determined. This can be done, for example, during the manufacture of the heat pump system 10, during the calibration of the heat pump system 10 or during the recalibration of the heat pump system 10. The system data 30 include at least input measurement data 12 and output measurement data 14. In addition, the system data 30 can also include external measurement data 20. Figure 1 In the example shown, in method step 24 the input measurement data 12 of the heat pump 36 are determined (see also Figure 4a The input measurement data 12 include, for example, the power request of the heat pump 36. Furthermore, in method step 24, the output measurement data 14 of the heat pump 36 are determined (see Figure 4c ). The output measurement data 14 include, for example, the generated (indirectly measured) heat flow. Furthermore, in method step 24, external measurement data 20 (see Figure 4bThe exogenous measurement data 20 include, for example, the temperature on the evaporator side of the heat pump 36. In this exemplary case, the determined input measurement data 12, the determined output measurement data 14, and the exogenous measurement data 20 form a data triple. The measurement values ​​of the data triple with the same index are each determined simultaneously.

[0044] In at least one further method step 26, a future input trajectory 28 of the heat pump system 10 is determined based on the system data 30 of the heat pump system 10 measured at an earlier time in method step 24. In method step 26, the future input trajectory 28 is determined using the target function defined in formula (5). In addition, in method step 26, a future output trajectory 38 and all other trajectories of the trajectory vector (except for possible exogenous input trajectories, for example, including external disturbances such as the outdoor temperature) are determined using the target function defined in formula (5). To this end, the target function is minimized / optimized. The determination of the future input trajectory 28 of the heat pump system 10 performed in method step 26 is achieved by means of data-driven predictive control of the heat pump system 10. The data-driven predictive control of the heat pump system 10 performed in method step 26 is limited to a first operating load range 16 and a second operating load range 18 of the heat pump 36. The first operating load range 16 and the second operating load range 18 do not overlap and are spaced apart from each other. The first operating load range 16 only includes zero operating load of the heat pump 36 (the heat pump 36 is in the off state). The second operating load range 18 includes a load range of the heat pump 36 in which normal operation is possible.

[0045] In method step 26, for the predictive regulation of the heat pump system 10, the solver of the calculation unit 22, 22' solves the regulation equation filled with system data 30. To this end, the solver minimizes the objective function of formula (5) while taking into account the predetermined limits / operating limits. The regulation equation is shown in formula (2) above. On one side of the equation, the regulation equation includes a measurement data matrix multiplied by a decision variable vector by matrix-vector multiplication, and on the other side of the equation, a trajectory vector with future trajectories 28, 38. The measurement data matrix is ​​filled with system data 30. The measurement data matrix is ​​composed of Hankel matrices of one measurement data type each including only one type of system data 30, which are stacked one above the other. The regulation equation includes binary variables. In order to take into account the lower operating limit of the heat pump system 10, the elements of the future input trajectory 28 and the initial value of the input trajectory in the regulation equation are modified by multiplying by binary variables. The elements of the future input trajectory 28 modified in this way each correspond to the prediction time step of the predicted input of the heat pump system 10. In method step 26, the solver optimizes the elements of the future input trajectory 28 to which the binary variables are assigned when solving the control equation using computing units 22, 22'. In method step 26, the solver optimizes the binary variables of the control equation, in particular the binary variables assigned to the elements of the input trajectory 28, when solving the control equation, preferably together with the elements of the future input trajectory 28. In method step 26, slack variables (if any) may also be optimized. In method step 26, the solver optimizes the decision variables of the decision variable vector of the control equation when solving the control equation. In method step 26, the solver optimizes at least the elements of the future output trajectory 38 when solving the control equation, the elements of which correspond to the time steps of the future input trajectory 28 to which the binary variables are assigned. In method step 26, the solver optimizes at least the future trajectory elements of the input trajectory 28 and the output trajectory 38 when solving the control equation. In method step 26, the solver does not change the elements of the measurement data matrix when solving the control equation. The elements of the measurement data matrix are invariant to the solver.

[0046] In substep 42 of method step 26, to simplify the solution of the control equation, in particular to simplify the minimization of the objective function, the elements of the future input trajectory 28 modified by the binary variable are each replaced by two inequalities. For example, the two inequalities shown in formula (3) each replace an element of the future input trajectory 28 in formula (2). The two inequalities are designed so that when the binary variable is 1, the values ​​permitted for solving the control equation for each corresponding element of the future input trajectory 28 are limited by two different, non-zero limit values, which represent an upper limit 56 and a lower limit 54, respectively. These limit values ​​define the permissible (input) operating range (partial load to full load) of the heat pump system 10. In addition, to achieve this task, the two inequalities are designed so that when the binary variable is 0, the values ​​permitted for solving the control equation for each corresponding element of the future input trajectory 28 are set to 0. This value of 0 corresponds to the deactivation of at least the heat pump 36 of the heat pump system 10, or the deactivation of the entire heat pump system, at the respective associated prediction time step.

[0047] In at least one further method step 48, the result of the first trajectory element of input trajectory 28 that can be assigned to a future time step of prediction horizon 40 is set on heat pump 36. In at least one further method step 50, the control equations are solved again using new initial values ​​(i.e., the values ​​set in method step 48).

[0048] Figure 3a An input power-time diagram 52 of a heat pump system 10 is shown. The relative input power of the heat pump 36 is plotted on the horizontal axis 62 of the input power-time diagram 52. Time is plotted in evenly spaced time steps on the vertical axis 64 of the power-time diagram 52. These time steps correspond to the number of time steps of the trajectory 28, 38 of the trajectory vector of the control equation. The total number of time steps for the data-driven predictive control results in a prediction range 40. In particular, the number of time steps, the prediction range 40 and the intervals between the time steps can be freely selected. The input trajectory 28 is recorded in the input power-time diagram 52. A first operating load range 16 is marked in the input power-time diagram 52. The first operating load range 16 comprises only the zero line of the input power-time diagram 52. In Figure 3a In the example shown, the input trajectory 28 is in the first operating load range 16 (δ k =0). The second operating load range 18 is marked in the input power-time graph 52. The second operating load range 18 includes a range between a lower limit 54 and an upper limit 56 of the relative input power. The lower limit 54 of the second operating load range 18 is at a relative input power of 0.2 (i.e., 20% of full load). The upper limit 56 of the second operating load range 18 is at a relative input power of 1 (i.e., full load). Figure 3a In the example shown, the input trajectory 28 is in the second operating load range 18 (δ k = 1). First operating load range 16 and second operating load range 18 do not overlap. First operating load range 16 and second operating load range 18 are spaced apart from each other. An exclusion range 58 exists between first operating load range 16 and second operating load range 18. Exclusion range 58 forces the two operating load ranges 16, 18 to be spaced apart from each other. Exclusion range 58 specifies the relative input power range within which heat pump 36 cannot operate. Input trajectory 28 determined using the method according to the present invention cannot have any values ​​within exclusion range 58. Lower limit 54 and upper limit 56 are selectable / variable.

[0049] Figure 3b A graph 60 of the output power of the heat pump system 10 is shown. The output power of the heat pump 36 is plotted on the horizontal axis 66 of the output power-time graph 60. Time is plotted in evenly spaced time steps on the vertical axis 68 of the power-time graph 52. These time steps correspond to the number of time steps of the trajectories 28, 38 of the control equation trajectory vector. The output trajectory 38 is recorded in the output power-time graph 60.

Claims

1. A method for data-driven predictive regulation of a heat pump system (10), in, determining at least one future input trajectory (28) of the heat pump system (10) based on at least system data (30) of the heat pump system (10) measured at an earlier time using a defined target function, in particular a defined target function to be minimized, The system data (30) includes at least input measurement data (12) and output measurement data (14) of the heat pump system (10). In this case, for data-driven predictive regulation of a heat pump system (10), a control equation filled with system data (30) is solved by means of a calculation unit (22) of the heat pump system (10) or by means of a calculation unit (22') arranged outside the heat pump system (10) and having a communication connection (70) with the heat pump system (10). The calculation unit (22, 22') determines a trajectory (28, 38) of the heat pump system (10), in particular at least the input trajectory (28) and at least one output trajectory (38) of the heat pump system (10), by means of a control equation based on a matrix-vector multiplication of at least one measurement data matrix and a decision variable vector. The measurement data matrix is ​​formed by at least two superimposed Hankel matrices, each of which only includes one measurement data type of the system data (30). It is characterized in that, in order to take into account the lower operating limit of the heat pump system (10), at least a plurality of elements of a future input trajectory (28) in a control equation are each assigned a binary variable, in particular by multiplication, wherein the elements of the future input trajectory (28) in particular each correspond to a predicted time step of a predicted input of the heat pump system (10).

2. The method according to claim 1, characterized in that When solving the control equations with the aid of a calculation unit (22, 22'), at least the elements of the future input trajectory (28) to which the binary variables are assigned are optimized, preferably at least the elements of the future input trajectory to which the binary variables are assigned are optimized together with the corresponding binary variables.

3. The method according to claim 2, characterized in that When solving the control equations with the aid of the calculation unit (22, 22'), at least the elements of the future output trajectory (38) are optimized, which elements correspond to time steps of the future input trajectory (28) to which the binary variables are respectively assigned.

4. The method according to any one of the preceding claims, characterized in that When solving the control equations with the aid of a calculation unit (22, 22'), at least the decision variables of the decision variable vector are optimized.

5. The method according to any one of the preceding claims, characterized in that When solving the control equations with the aid of the calculation unit (22, 22'), the elements of the measurement data matrix remain unchanged.

6. The method according to any one of the preceding claims, characterized in that To solve the control equations by means of the calculation units (22, 22'), elements of the trajectory (28, 38) modified by binary variables, in particular elements of the future input trajectory (28) modified by binary variables, are replaced by two inequalities.

7. The method according to claim 6, characterized in that The two inequalities are designed such that, when the binary variable is 1, the values ​​permitted for solving the control equation by means of the calculation unit (22, 22') for the respective associated elements of the trajectory (28, 38), in particular the respective associated elements of the future input trajectory (28), are limited by two respectively different and non-zero limit values, in particular an upper limit (56) and a lower limit (54).

8. The method according to claim 7, characterized in that The limit values ​​correspond to the permissible (input) operating range (part load to full load) of the heat pump system (10).

9. The method according to any one of claims 6 to 8, characterized in that The two inequalities are designed such that when the binary variable is 0, the values ​​of the associated elements of the trajectory (28, 38), in particular the associated elements of the future input trajectory (28), which are permitted for solving the control equation by means of the calculation unit (22, 22'), are set to the value 0.

10. The method according to claim 9, characterized in that The value 0 corresponds to the deactivation of at least one heat pump (36) of the heat pump system (10), in particular in the respectively associated prediction time step.

11. A computing unit (22, 22') configured for data-driven predictive regulation of a heat pump system (10) using the method according to any one of claims 1 to 10.

12. A heat pump system (10), comprising at least one heat pump (36) and at least one computing unit (22, 22') according to claim 11, wherein the computing unit is integrated into the heat pump system (10) or is arranged outside the heat pump system (10) and has a communication connection (70) with the heat pump system (10).

Images