Predictive control method based on improved locust optimization

Abstract

The invention provides a prediction control method based on improved locust optimization, and relates to the technical field of data processing and control engineering, and the method comprises the steps: carrying out the parameter identification of a mass regulation channel of a heat exchange station system through employing a recursive least square method with a forgetting factor, and building a controlled autoregressive moving average integral model; based on the controlled autoregressive moving average integral model, generating a prediction output through implicit generalized prediction control and constructing a target function, the implicit generalized prediction control directly identifying a prediction matrix and a prediction vector; when the control quantity increment exceeds a preset constraint range, an improved locust optimization algorithm is adopted to optimize the target function, the improved locust optimization algorithm is combined with a double-differential evolution strategy and a reverse learning mechanism, the double-differential evolution strategy updates the locust position through differential variation, and the locust position is updated through differential variation; the variant basis vector is based on the current optimal location and the random individual location.

Description

Technical Field

[0001] The present application belongs to the field of data processing and control engineering technology, and specifically relates to a prediction control method based on improved locust optimization. Background Art

[0002] With the improvement of industrial automation level, especially the development of emerging technologies such as Internet of Things, artificial intelligence and cloud computing, the demand for precise control of complex systems is increasing. As an important part of the heating system, the control strategy of the heat exchange station directly affects the energy efficiency and stability of the system. Traditionally, the heat exchange station mainly relies on PID controller for adjustment, but when faced with complex dynamic environments and nonlinear characteristics, PID controllers often show problems such as slow response and large overshoot.

[0003] In order to overcome these limitations, researchers have proposed a variety of improvement schemes, including but not limited to fuzzy control, adaptive control, internal model control, and generalized predictive control (GPC) based on particle swarm optimization. Fuzzy control improves the performance of PID by introducing expert knowledge, but the rule design is more complicated; adaptive control improves the robustness of the system, but it is highly sensitive to parameter changes and difficult to adjust. Although internal model control provides good control effects in theory, it faces the challenge of model uncertainty in practical applications. Although particle swarm optimization combined with GPC improves the accuracy of output prediction, it is easy to fall into the problem of local optimal solution, which affects the overall control performance.

[0004] The main disadvantages of the existing technology are: first, most advanced control algorithms have high computational complexity, resulting in insufficient real-time performance; second, when dealing with large lag and nonlinear problems, traditional methods are difficult to achieve ideal control accuracy; third, although some intelligent optimization algorithms such as ant colony algorithm and particle swarm optimization are used to improve control effects, due to the limitations of the search mechanism, they still have deficiencies in global search capabilities and avoiding local optimality. Summary of the invention

[0005] The present application provides a predictive control method based on improved locust optimization to solve the technical problems in the prior art of low computational efficiency, easy to fall into local optimal solution and insufficient control accuracy of the heat exchange station control system when facing a complex dynamic environment.

[0006] The technical solution adopted in this application is: The present application embodiment provides a prediction control method based on improved locust optimization, comprising: S1: The recursive least square method with forgetting factor is used to identify the parameters of the mass regulation channel of the heat exchange station system, and a controlled autoregressive moving average integral model is established; S2: Based on the controlled autoregressive moving average integral model, generating a predicted output and constructing an objective function through implicit generalized predictive control, wherein the implicit generalized predictive control directly identifies a prediction matrix and a prediction vector; S3: When the control amount increment exceeds the preset constraint range, the objective function is optimized by using an improved locust optimization algorithm, wherein the improved locust optimization algorithm combines a double differential evolution strategy and a reverse learning mechanism, wherein: The double differential evolution strategy updates the locust position through differential mutation, and the mutation basis vector is based on the current optimal position and the random individual position; The reverse learning mechanism updates the locust position by comparing the fitness value of the current position of the locust individual with the reverse position; S4: adjusting the control amount according to the optimization result to achieve the optimal control of the heat exchange station system.

[0007] According to one embodiment of the present application, in step S1, the expression of the controlled autoregressive moving average integral model is:

[0008] in, It is a differential operator, the input u(k) is the frequency conversion of the primary booster pump, and the output y(k) is the secondary supply and return water temperature difference.

[0009] According to one embodiment of the present application, in step S2, the expression of the objective function is:

[0010] Among them, n is the prediction time domain, m is the control time domain, λ k is the control weighting constant.

[0011] According to one embodiment of the present application, in step S3, the locust position update formula of the double differential evolution strategy is:

[0012] in, x i (t) is the position of the i-th locust in the iteration process, x j (t) is the random position of the locust, The optimal position of , F is the scaling factor.

[0013] According to one embodiment of the present application, the updating formula of the scaling factor F is:

[0014] Among them, φ∈[0,T] is a constant.

[0015] According to one embodiment of the present application, in step S3, the locust position update formula of the reverse learning mechanism is:

[0016] in, x i d (t) is the d-dimension of the i-th locust in the t-th iteration, ub d , lb d is the boundary of the locust in the dth dimension, and rand is a random number.

[0017] According to one embodiment of the present application, in step S3, γ% of individuals are randomly selected from the initial population, with the current control amount increment as the initial value, and the remaining individuals are randomly generated within the constraint range.

[0018] According to one embodiment of the present application, in step S4, the constraint condition of the control amount increment is:

[0019] in, is the constraint vector, and To control the range of increments, define ; and To control the minimum and maximum values ​​of the increment.

[0020] A computer-readable storage medium stores a program, which, when executed by a processor, implements the steps in a predictive control method based on improved locust optimization.

[0021] An electronic device comprises a memory, a processor and a program stored in the memory and executable on the processor, wherein when the processor executes the program, steps of a predictive control method based on improved locust optimization are implemented.

[0022] Due to the adoption of the above technical solution, the beneficial effects achieved by this application are as follows: This application uses the recursive least squares method with forgetting factor (FFRLS) for model identification, which can effectively deal with data saturation problems, quickly adapt to changes in system parameters, and establish an accurate controlled autoregressive moving average integral model (CARIMA), providing a solid foundation for subsequent control strategies. Compared with traditional PID control and other advanced control methods, this application reduces computational complexity while improving model accuracy.

[0023] Secondly, the application of implicit generalized predictive control (IGPC) further simplifies the calculation process, directly identifying G and f from the input and output data, avoiding the problem of too long time caused by the traditional GPC based on the Diophantine equation solution, thereby speeding up the response speed and enhancing the real-time performance of the system. In addition, the integration of the double differential evolution strategy and the reverse learning mechanism into the locust optimization algorithm greatly improves the global search capability and local development efficiency, ensuring that the algorithm can find the optimal solution in a short time and avoid the risk of falling into the local optimum. Simulation experiments show that under the same disturbance conditions, the strategy proposed in this application not only shortens the adjustment time and reduces the overshoot compared with other methods, but also significantly reduces the disturbance recovery time, improving the robustness and stability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings: Figure 1 A flow chart of a predictive control method based on improved locust optimization provided in an embodiment of the present application. DETAILED DESCRIPTION

[0025] In order to more clearly illustrate the overall concept of the present application, a detailed description is given below in an illustrative manner in conjunction with the accompanying drawings.

[0026] In the following description, many specific details are set forth to facilitate a full understanding of the present application. However, the present application may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present application is not limited by the specific embodiments disclosed below. It should be noted that the embodiments of the present application and the features in each embodiment may be combined with each other without conflict.

[0027] In the present application, unless otherwise clearly specified and limited, a first feature "above" or "below" a second feature may be that the first and second features are in direct contact, or the first and second features are in indirect contact through an intermediate medium. In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in an appropriate manner in any one or more embodiments or examples.

[0028] like Figure 1 As shown, a predictive control method based on improved locust optimization comprises: S1: The Forgetting Factor Recursive Least Squares (FFRLS) method is used to identify the parameters of the quality regulation channel of the heat exchange station system, and a Controlled Auto-Regressive and Moving Average with Integrative (CARIMA) model is established.

[0029] Specifically, Recursive Least Squares (RLS): This is an online estimation method used to update system parameter estimates in real time, and is particularly suitable for dynamically changing environments.

[0030] Forgetting factor: The introduction of the forgetting factor (λ) in FFRLS enables the algorithm to give higher weight to new data while gradually "forgetting" old data, thereby better adapting to changes in system parameters over time.

[0031] First, we select the initial parameter vector θ(0) and the covariance matrix P(0). Typically, these initial values ​​can be set to zero or small values ​​selected empirically.

[0032] Collect actual operation data from the heat exchange station system, including input (such as booster pump frequency conversion u(k)) and output (such as supply and return water temperature difference y(k)). Since the quality regulation channel of the heat exchange station system can be described by a second-order plus lag link, it is necessary to ensure that the collected data can fully reflect this characteristic.

[0033] For each sampling time k, the parameter estimates are updated using the FFRLS algorithm: Calculate the prediction error at the current moment

[0034] in, is the regression vector.

[0035] Update gain vector

[0036] Update parameter vector

[0037] Update the covariance matrix

[0038] After completing the above parameter identification, the CARIMA model is established based on the obtained parameter estimates. The form of the model is:

[0039] in, , and are the polynomials for the autoregressive, moving average, and noise models, respectively; represents the difference operation, which is used to eliminate non-stationarity; d is the number of delay steps.

[0040] In the heat exchange station system, the input u(k) represents the frequency conversion of the primary network valve, and the output y(k) represents the supply and return water temperature difference of the secondary network. In this way, not only can the dynamic behavior of the system be accurately captured, but also the data saturation problem can be effectively handled, thus improving the accuracy and robustness of the model.

[0041] For example, suppose a heat exchange station system is being optimized. One of the key indicators of the system is the secondary network supply and return water temperature difference y(k), which is mainly affected by the primary network booster pump frequency conversion u(k). The goal is to identify the system parameters through the FFRLS method and establish an accurate CARIMA model based on this for subsequent control strategy design.

[0042] First, it is necessary to collect actual operating data from the heat exchange station system over a period of time. For example, the data of the booster pump frequency conversion u(k) and the supply and return water temperature difference y(k) can be recorded every 5 minutes in a day. Assume that the following data points are collected: u(1),y(1) u(2),y(2) ... u(N),y(N) Initialize the FFRLS algorithm Next, you need to initialize the relevant parameters of the FFRLS algorithm: Set the initial estimated parameter vector θ(0) to be the zero vector.

[0043] The covariance matrix P(0) is set to a large positive definite matrix, such as , where I is the identity matrix.

[0044] The forgetting factor λ is usually set between 0.95 and 0.99, and here λ=0.98 is taken.

[0045] For each sampling time instant k (from 1 to N), perform the following steps: Calculate the prediction error: Based on the current parameter estimates and input and output data , calculate the prediction error

[0046] Update gain vector

[0047] Update parameter vector

[0048] Update the covariance matrix

[0049] After the above FFRLS iterations, a set of optimal parameter estimates θ are obtained. Using these parameters, a CARIMA model can be constructed to describe the dynamic behavior of the heat exchange station system. The specific form is as follows:

[0050] In this example, since we are concerned about the effect of the booster pump frequency conversion u(k) on the supply and return water temperature difference y(k), it can be simplified as:

[0051] Where Δ represents the differential operation , It is a parameter estimation value obtained through FFRLS, which represents the degree of influence of booster pump frequency conversion on the supply and return water temperature difference.

[0052] In this way, the parameters of the mass regulation channel of the heat exchange station system were identified using the FFRLS method, and an accurate CARIMA model was established, which provided a solid foundation for the further implementation of implicit generalized predictive control. This model can not only accurately reflect the dynamic characteristics of the system, but also effectively deal with the data saturation problem and improve the performance of the overall control strategy.

[0053] Furthermore, real-time data fusion can be performed: in addition to using data from a fixed time period for initial parameter identification, a mechanism can be designed to integrate new operating data online to achieve continuous updating of model parameters. This will enable the model to better adapt to parameter drift caused by environmental changes or system aging. Furthermore, dynamic forgetting factor adjustment can be performed: the value of the forgetting factor λ can be automatically adjusted according to changes in the system state. For example, when the system is subject to a large disturbance, λ can be reduced to respond to new situations more quickly; and λ can be increased during the stable period to reduce dependence on historical data.

[0054] S2: Based on the controlled autoregressive moving average integral model, the prediction output is generated and the objective function is constructed through implicit generalized predictive control, and the implicit generalized predictive control is achieved by directly identifying the prediction matrix and the prediction vector.

[0055] Specifically, firstly, the recursive least squares method with forgetting factor (FFRLS) has been used to identify the parameters of the mass regulation channel of the heat exchange station system, and a CARIMA model has been established. This model is used to describe the behavior of the system, and its general form is:

[0056] , and are the polynomials for the autoregressive, moving average, and noise models, respectively; represents the differential operation, which is used to eliminate non-stationarity; d is the number of delay steps, u(k) is the input (such as the frequency conversion of the booster pump), and y(k) is the output (such as the supply and return water temperature difference).

[0057] In traditional generalized predictive control (GPC), it is usually necessary to use Diophantine equations to calculate the output value at future moments, which involves complex matrix operations and takes a long time. The IGPC method simplifies this process by directly identifying the prediction matrix G and the prediction vector f, thereby improving computational efficiency.

[0058] In the heat exchange station system, the prediction model of the quality adjustment channel adopts the CARIMA model, which is expressed as follows:

[0059] in, It is a differential operator, the input u(k) is the frequency conversion of the booster pump, and the output y(k) is the supply and return water temperature difference.

[0060] In order to improve the robustness of the system, the influence of the current primary network valve frequency conversion u(k) on the future secondary network supply and return water temperature difference y(k) is taken into consideration in the objective function, and the objective function can be defined as:

[0061] in, To predict the output, is the reference trajectory used to guide the transition process. is the control weight constant, balancing the relationship between control action and tracking error. N is the prediction time domain, and M is the control time domain.

[0062] Traditional GPC solves the output value at the future k+j time based on the Diophantine equation, but this method is computationally complex and time-consuming. Therefore, IGPC adopts a direct identification method, using the primary network booster pump valve frequency conversion input by the heat exchange station system and the output secondary network temperature difference data to directly identify the prediction matrix G and prediction vector f.

[0063] The specific steps are as follows: Solve the matrix G: Write n parallel predictors based on the GPC optimal output prediction matrix. The last row of equations contains all the elements of the matrix G within the optimal control rate Δu. Therefore, we only need to calculate the y(k+n) equation to get G.

[0064] Solve the prediction vector f: Write the f at time k+1 from the n-step estimate at time k,

[0065] in, is the weight factor, is the prediction error.

[0066] Due to the restrictions on the process parameter range and adjustment frequency of the heat exchange station control system, the calculation of the control quantity increment Δu in the rolling optimization link is subject to certain interval constraints. The constraint range is defined as follows based on the actual working conditions:

[0067] When the calculated optimal control rate does not satisfy these constraints, further optimization is required.

[0068] Based on the above prediction model and rolling optimization strategy, an objective function is constructed to achieve optimal control. The objective function not only considers the deviation between the system output and the reference trajectory, but also considers the cost of the control action to ensure that the system can respond quickly and maintain stable operation.

[0069] For example, firstly, the recursive least square method with forgetting factor (FFRLS) has been used to identify the parameters of the heat exchange station system, and a CARIMA model describing the system behavior has been established. This model can accurately reflect the relationship between the primary network booster pump frequency conversion (input u(k)) and the secondary network supply and return water temperature difference (output y(k)).

[0070] Next, the IGPC method will be used to optimize the control strategy. The specific steps are as follows: Data was collected over a period of time from the actual operation of the heat exchange station, including the booster pump variable frequency setting value and the corresponding supply and return water temperature difference changes.

[0071] This data is used to validate and tune the model, ensuring it can accurately predict future system behavior.

[0072] At each time point k, it is necessary to predict the supply and return water temperature difference changes in the future (e.g., the next 5 minutes, 10 minutes, etc.). The key here is "direct identification", that is, instead of calculating these predicted values ​​through complex mathematical operations, the prediction matrix G and prediction vector f are directly estimated using the previously established CARIMA model.

[0073] In other words, by analyzing historical data and current status, we can directly get the expected output value in the next few time steps. This makes the prediction process more efficient and reduces the computational burden.

[0074] The objective function is designed to minimize the difference between the actual output and the desired output, taking into account the cost of the control action. For example, if the goal is to keep the supply and return temperature difference stable within a certain range, then the objective function will penalize actions that cause the temperature to deviate from the set point.

[0075] In addition, the objective function also considers the cost of the control action itself. For example, frequent adjustment of the booster pump frequency conversion may increase equipment wear or energy consumption. Therefore, when constructing the objective function, these two factors will be considered comprehensively to find a balance point.

[0076] Assume that k=0 at the current moment, and hope to maintain the supply and return water temperature difference near a certain ideal value in the next 10 minutes. Based on historical data and current status, use the IGPC method: Predict future output: Extract information directly from the existing CARIMA model and predict the supply and return water temperature difference trend every minute in the next 10 minutes.

[0077] Adjust the control strategy: Based on these predicted values, calculate the specific adjustments that need to be made to the booster pump frequency conversion to minimize the difference between the actual output and the ideal value. In this process, not only pay attention to the final result, but also pay attention to avoiding the negative effects of over-adjustment.

[0078] Monitoring and feedback: After implementing a new control strategy, continuously monitor system performance and dynamically adjust model parameters or control strategies based on actual conditions to ensure that the system is always in the best condition.

[0079] Furthermore, a multi-input multi-output (MIMO) system can be used: for complex heat exchange station systems with multiple input and output variables, the single-variable IGPC method can be extended to a multi-variable form, while considering the mutual influence between different variables to build a more comprehensive and accurate system model. Furthermore, nonlinear expansion can be used: considering that the actual heat exchange station system may have significant nonlinear characteristics, nonlinear elements such as neural networks or fuzzy logic can be introduced on the basis of the existing linear CARIMA model to form a hybrid model to improve prediction accuracy.

[0080] S3: When the control amount increment exceeds the preset constraint range, the objective function is optimized by using an improved locust optimization algorithm, wherein the improved locust optimization algorithm combines a double differential evolution strategy and a reverse learning mechanism, wherein: The double differential evolution strategy updates the locust position through differential mutation, and the mutation basis vector is based on the current optimal position and the random individual position; The reverse learning mechanism updates the locust position by comparing the fitness value of the current position of the locust individual with the reverse position.

[0081] Specifically, in the optimization control of the heat exchange station system, when the calculated control quantity increment (such as the change in the frequency conversion of the booster pump) exceeds the preset constraint range, the improved locust optimization algorithm is used to optimize the objective function. This method combines the double differential evolution strategy and the reverse learning mechanism to enhance the global search capability and avoid the problem of local optimal solution. The following is a detailed explanation: In a control system, in order to ensure the stability and safety of the system, a certain constraint range is usually set for the control quantity increment, that is, the minimum value and maximum value If the control quantity increment calculated according to the current control strategy exceeds this range, the increment needs to be further optimized.

[0082] Differential mutation to update locust positions: In the traditional locust optimization algorithm, the locust position is updated based on the best position in the group. In the improved method, a double differential evolution strategy is introduced to update the locust position.

[0083] Selection of basis vectors: In this process, the mutation basis vector not only considers the optimal position in the current population, but also combines a randomly selected individual position. The purpose of this is to increase the diversity of the population and thus improve the global search capability.

[0084] Specific operation: Assumption represents the position of the i-th locust at iteration t, is another randomly selected locust position, and Represents the optimal position in the current population. The difference between these two positions is used to generate a new position update direction, and a scaling factor F is used to adjust the step size in this direction. This allows the algorithm to explore the solution space more flexibly.

[0085] Comparing the fitness values ​​of the current position and the reverse position: The reverse learning mechanism is a technology used to enhance local development capabilities. It compares the fitness values ​​of the current position of each locust individual and its reverse position, and selects the better position for update.

[0086] The concept of reverse position: For each locust individual position , its inverse position can be defined as some form of mapping, usually reflection based on boundary conditions. For example, in d-dimensional space, if and are the upper and lower bounds of the dimension respectively, then the reverse position can be obtained by the formula to calculate.

[0087] Best update: If the reverse position is found to have a better fitness value, the new position is used to replace the original position and brought into the next iteration; otherwise, the current position is continued to be locally developed to find a better solution. This mechanism helps prevent the algorithm from converging to the local optimum too early and increases the possibility of finding the global optimal solution.

[0088] For example, suppose that the frequency conversion of the primary network booster pump of the heat exchange station system is being adjusted to optimize the temperature difference between the secondary network supply and return water. During a rolling optimization process, the calculated control quantity increment exceeds the preset constraint range. At this time, the improved locust optimization algorithm is started: First, a double differential evolution strategy is used to generate a new set of potential solutions, exploring new areas of the solution space by combining the current optimal solution and the position differences of random individuals.

[0089] Then, a reverse learning mechanism is applied to examine the performance of each locust individual and its reverse position, and the position with better performance is selected as the starting point for the next generation.

[0090] Finally, after many iterations, the control increment that satisfies the constraints and optimizes the objective function is found.

[0091] In this way, the improved locust optimization algorithm effectively solves the problem that traditional optimization methods are prone to fall into local optimality, and improves the control accuracy and response speed of the entire system.

[0092] For example, suppose that the frequency conversion u(k) of the primary network booster pump of a heat exchange station system is being optimized to achieve the ideal secondary network supply and return water temperature difference y(k). Through implicit generalized predictive control (IGPC), a control quantity increment Δu has been calculated, but this increment exceeds the preset minimum value. and maximum value At this time, it is necessary to use the improved locust optimization algorithm to find the optimal control quantity increment so that the objective function reaches the optimal value while satisfying the constraints.

[0093] Set initial parameters: First, initialize the locust population, including population size, number of iterations, etc. Assume there are 50 locusts, each of which represents a potential incremental solution for the control quantity.

[0094] Define the objective function: The objective function can be a combination of performance indicators such as settling time, overshoot, or ITAE indicators. For example, the goal is to minimize the settling time and overshoot while ensuring that the disturbance recovery time is as short as possible.

[0095] Mutation basis vector generation: For each locust individual, select the optimal position in the current population and another randomly selected individual position As the basis vector. Assume that the current optimal position is , the random individual position is .

[0096] Specific operation: For the position of the i-th locust , according to the formula:

[0097] Where F is a scaling factor, usually set to a positive number less than 1 (such as 0.5). This generates a new position , used to explore new areas of the solution space.

[0098] Update locust position: the new position where it will be generated With original position Compare and select the position with better fitness as the basis for the next iteration.

[0099] Calculate the reverse position: For each locust individual position , calculate its reverse position. Assume that in two-dimensional space, the upper and lower bounds are and Then the reverse position can be obtained by the formula:

[0100] For example, if the current position is , then the reverse position is

[0101] Best update: compare current positions and reverse position If the reverse position has a better fitness, it is used to replace the original position and brought into the next iteration; otherwise, the current position is continued to be locally developed.

[0102] Multiple iterations: Repeat the above double differential evolution strategy and reverse learning mechanism process until the predetermined maximum number of iterations is reached or the control increment that satisfies the constraints and makes the objective function optimal is found.

[0103] For example: Suppose after several iterations, a control increment is found , which not only satisfies the preset constraints (i.e. ), and makes the objective function (such as adjustment time, overshoot, etc.) reach the optimal value.

[0104] Finally, the optimal control increment obtained by applying the improved locust optimization algorithm was applied to the actual operation of the heat exchange station system. For example, when adjusting the booster pump frequency conversion, using this new control increment can significantly reduce the adjustment time and overshoot, while speeding up the system response speed and improving the overall control effect.

[0105] Furthermore, abnormality detection and diagnosis can be performed: using established accurate models to analyze data and identify abnormal behaviors or potential failure points in the system. For example, statistical analysis of residuals (the difference between actual output and model prediction) can be used to detect deviations from the normal range. Furthermore, predictive maintenance can be performed: based on long-term accumulated data and model prediction capabilities, early warning of possible problems can be given, and preventive maintenance plans can be developed to reduce the risk of sudden failures and extend the service life of equipment.

[0106] S4: adjusting the control amount according to the optimization result to achieve the optimal control of the heat exchange station system.

[0107] Specifically, adjusting the control quantity according to the optimization result to achieve the optimal control of the heat exchange station system means that after completing the optimization process of the objective function, the actual control parameters in the system (such as the booster pump frequency conversion) are adjusted according to the obtained optimal control quantity increment to achieve the optimal system performance. The following is a detailed explanation: After optimizing the objective function using the improved locust optimization algorithm, a set of optimal control increment values ​​are obtained. These values ​​are obtained through a series of iterative calculations, aiming to minimize the objective function while satisfying all preset constraints. The objective function may include performance indicators such as adjustment time, overshoot, and disturbance recovery time.

[0108] Once the optimal control increment is obtained, the next step is to apply it to the actual control system. The specific steps are as follows: Determine the current status: First confirm the current operating status of the heat exchange station system, including the current booster pump frequency conversion settings and other related parameters.

[0109] Calculate the new control quantity: Based on the optimization results, calculate the specific value that needs to be adjusted. For example, if the current booster pump frequency conversion setting is 50Hz, and the optimization results show that it needs to be increased by 3Hz, the new booster pump frequency conversion setting should be 53Hz.

[0110] Implement the adjustment: Apply the calculated new control quantity to the system. This usually involves sending instructions to the control system to change the frequency setting of the booster pump. During this process, ensure that the adjustment is made within a safe range and will not cause equipment overload or other potential problems.

[0111] After adjusting the control amount, it is very important to closely monitor the response of the system: Real-time monitoring: Continuously observe changes in key performance indicators, such as whether the supply and return water temperature difference is close to the set value, whether there is obvious overshoot or oscillation, etc.

[0112] Feedback mechanism: If any deviation from the expected situation occurs, corrective measures can be taken quickly. For example, if the temperature change is found to be too drastic, it can be further optimized by fine-tuning the control amount.

[0113] In order to maintain the long-term efficient operation of the system, it is necessary to repeat the above optimization process regularly or irregularly: Periodic evaluation: As time goes by and environmental conditions change, the original optimization plan may no longer be applicable. Therefore, it is recommended to re-evaluate system performance regularly and optimize again based on the latest data.

[0114] Dynamic adjustment: In some cases, an online learning mechanism can also be designed to enable the system to automatically adjust itself based on the latest operating data, thereby achieving truly intelligent control.

[0115] In this way, adjusting the control quantity according to the optimization results not only improves the immediate performance of the heat exchange station system, but also enhances its ability to cope with various working conditions, ensuring the stability and reliability of the system. This method enables the heat exchange station to always maintain efficient operation under different working conditions, reducing energy consumption and extending the service life of the equipment.

[0116] For example, in a typical heat exchange station system, it is necessary to adjust the frequency conversion u(k) of the primary network booster pump to control the secondary network supply and return water temperature difference y(k) to achieve the best heating effect. An optimal control quantity increment Δu is calculated by implicit generalized predictive control (IGPC) combined with the improved locust optimization algorithm, and now this increment needs to be applied to the actual control system.

[0117] Assume that after a series of iterative calculations, the improved locust optimization algorithm gives the following optimization results: The optimal control quantity increment Δu=0.8 (assuming the unit is Hz), which means that the current booster pump frequency conversion needs to be increased by 0.8Hz. This increment is obtained while minimizing system performance indicators such as adjustment time, overshoot, and disturbance recovery time.

[0118] Before making any adjustments, first confirm the current state of the system: the booster pump frequency conversion setting is currently 50Hz.

[0119] The actual measured secondary network supply and return water temperature difference is 5°C, while the target value is 6°C.

[0120] Based on the optimization results, the new booster pump frequency conversion setting is calculated: new booster pump frequency conversion setting = current setting + control amount increment = 50Hz + 0.8Hz = 50.8Hz.

[0121] Next, the calculated new control quantity is applied to the actual control system: Send instructions: Send instructions through the control system to adjust the booster pump frequency from 50Hz to 50.8Hz.

[0122] Execute adjustment: After receiving the command, the control system automatically adjusts the operating frequency of the booster pump to 50.8Hz.

[0123] After the adjustment is completed, closely monitor the system's response: Real-time monitoring: Continuously observe changes in key performance indicators. For example, check whether the supply and return water temperature difference is gradually approaching the target value of 6°C.

[0124] Record data: Record the adjusted parameters, including the new supply and return water temperature difference, system response time and any possible overshoot or oscillation.

[0125] Assume that within a few minutes after the adjustment, the supply and return water temperature difference begins to gradually increase and eventually stabilizes at around 6°C, indicating that the system has achieved the expected effect. At the same time, there is no obvious overshoot or oscillation, indicating that the adjustment is successful.

[0126] If you find that the system performance is not completely as expected, you can take the following measures to fine-tune it: Feedback mechanism: If the supply and return water temperature difference exceeds the set target value (such as reaching 6.2°C), fine-tuning can be performed by slightly reducing the booster pump frequency (for example, reducing it by 0.1Hz).

[0127] Continuous optimization: Regularly re-evaluate system performance and optimize again based on the latest operating data. This helps ensure long-term efficient operation of the system under different operating conditions.

[0128] In this example, the optimal control increment (0.8Hz) was found through the improved locust optimization algorithm and applied to the variable frequency control of the booster pump of the heat exchange station system. The specific steps include determining the current state, calculating the new control amount, implementing adjustments, and monitoring the system response. Finally, through these steps, the optimal control of the heat exchange station system was achieved, so that the supply and return water temperature difference reached the expected target value, and the system maintained good stability and responsiveness.

[0129] In some embodiments of the present application, in step S1, the expression of the controlled autoregressive moving average integral model is:

[0130] in, It is a differential operator, the input u(k) is the frequency conversion of the primary booster pump, and the output y(k) is the secondary supply and return water temperature difference.

[0131] Specifically, the difference operator: In this context, the difference operator Δ is used to deal with trend components and non-stationarity issues in the data. In simple terms, it helps to eliminate long-term trends in the data, making the processed data more stable and stationary. This step is crucial to building an accurate model because it can reduce prediction errors and improve the accuracy of the model.

[0132] Input u(k): refers to the frequency conversion setting value of the primary network booster pump in the heat exchange station system. The function of the booster pump is to adjust the speed and pressure of the water flow, thereby affecting the efficiency of the entire heating system. Therefore, the change of u(k) directly affects the operating status of the system.

[0133] Output y(k): refers to the temperature difference between the supply and return water of the secondary network. This is a key indicator to measure the performance of the heat exchange station system, reflecting the efficiency of the system in transferring heat from the primary network to the secondary network. Ideally, it is hoped that by adjusting u(k), y(k) will be as close to the set target value as possible.

[0134] Purpose: By building a CARIMA model, the goal is to find a way to predict future output (i.e., supply and return water temperature difference) based on current and past inputs (i.e., booster pump frequency conversion settings). This predictive capability is critical to achieving precise control because it allows more scientific and reasonable decisions to be made in actual operations, thereby optimizing system performance.

[0135] Parameter Identification: In this step, the parameters of the CARIMA model are estimated using the Forgetting Factor Recursive Least Squares (FFRLS) method. These parameters describe the internal working mechanism of the system, including how the various variables interact with each other. By accurately estimating these parameters, the dynamic behavior of the system can be better understood and effective control strategies can be formulated accordingly.

[0136] In some embodiments of the present application, in step S2, the expression of the objective function is:

[0137] Among them, n is the prediction time domain, m is the control time domain, is the control weighting constant.

[0138] Specifically, the objective function is a key tool used in the optimization process to evaluate the pros and cons of different control strategies. In this particular application scenario, the objective function is designed to minimize the difference between the system output and the reference trajectory while taking into account the cost of the control action. In this way, the objective function helps to ensure system performance while also taking into account the actual cost of the control action (such as energy consumption, equipment wear, etc.).

[0139] Prediction time domain: refers to how long in the future you want to predict the system behavior. For example, if the prediction time domain is set to 10 minutes, then the objective function will consider the expected performance of the system in the next 10 minutes from the current moment.

[0140] Purpose: A longer prediction horizon can provide more comprehensive future information and help make more forward-looking decisions; however, a prediction horizon that is too long may increase computational complexity and affect real-time performance.

[0141] Control time domain: refers to the time in the future when the control measures are planned to be taken. For example, if the control time domain is set to 5 minutes, it means that specific control actions will be formulated within the next 5 minutes.

[0142] Purpose: A shorter control time domain can respond to system changes more quickly and improve the system's response speed; however, if the control time domain is too short, long-term effects may not be fully taken into account, resulting in frequent adjustments to the control amount and increased system burden.

[0143] Control weighting constant Control weighting constant: It is used to balance the relationship between the system output error and the cost of the control action. Specifically, it determines the weight distribution between the need for the system output to be close to the reference trajectory and the need to reduce the amplitude of the control action during the optimization process.

[0144] Purpose: By adjusting The value of can find a suitable balance between pursuing high-precision control and reducing control costs. The value tends to reduce the control action range, thereby reducing energy consumption or equipment wear; while lower The value focuses more on making the system output as close to the reference trajectory as possible.

[0145] Assume that you are operating a heat exchange station system and want to optimize the booster pump variable frequency setting through the IGPC method to achieve the ideal supply and return water temperature difference: Setting parameters: Select prediction time domain n = 10 minutes, which means that the supply and return water temperature difference changes in the next 10 minutes will be predicted. Control time domain m = 5 minutes, which means that specific control actions will be implemented during this period. Control weighted constant Set to an appropriate value, such as 0.1, to balance output error and control cost.

[0146] Calculate the objective function value: Based on the current system status and the prediction model, the expected supply and return water temperature difference is calculated every minute in the next 10 minutes and compared with the set target value.

[0147] At the same time, consider the control actions that need to be taken within these 5 minutes (such as adjusting the frequency conversion of the booster pump) and calculate the cost of these actions.

[0148] Optimize control action: Based on the above calculation results, determine the optimal booster pump frequency conversion setting to minimize the objective function value. This usually means finding a solution that can make the supply and return water temperature difference as close to the target value as possible while keeping the control action cost as low as possible.

[0149] Implement and monitor: Apply the calculated optimal control quantity to the actual system and continuously monitor the system’s response to ensure that the system operates as expected.

[0150] In some embodiments of the present application, in step S3, the locust position update formula of the double differential evolution strategy is:

[0151] in, is the position of the i-th locust in the iteration process, is the random position of the locust, The optimal position of , F is the scaling factor.

[0152] Specifically, the double differential evolution strategy is a method that combines the idea of ​​differential evolution (DE) to improve the exploration ability of the traditional locust optimization algorithm. It guides the locust colony to move towards a better solution by generating new position candidates. This method is particularly suitable for complex and dynamically changing optimization problems.

[0153] Locust positions: Each locust represents a potential solution or control volume increment. At each iteration, these positions need to be updated in order to find a better solution.

[0154] The position of the i-th locust during the iteration ( ): This refers to the position of the i-th locust at the current iteration t. The position of each locust represents a set of possible control parameter values.

[0155] Another locust position is randomly selected ( ): In order to increase population diversity and explore more areas of the solution space, the position of another locust is randomly selected as a reference point when updating the position.

[0156] Current best position ( ): The best currently known position in the entire population. This position corresponds to the best solution or lowest objective function value found so far.

[0157] Differential mutation: In this strategy, instead of simply updating based on the position of the individual itself or other single individuals, a new position is generated by calculating the difference between the positions of two individuals. Specifically, the difference between the current optimal position and the position of a randomly selected individual is considered, and this difference is used to guide the position update of other individuals.

[0158] Scaling factor (F): This is an important parameter used to adjust the step size of differential mutation. The scaling factor determines the magnitude of the change in the new position relative to the original position. A smaller scaling factor results in a smaller step size, making the search more refined; a larger scaling factor allows for larger jumps, which helps to jump out of the local optimal solution faster.

[0159] Assume that the frequency conversion setting of the primary network booster pump of the heat exchange station system is being optimized to achieve the ideal secondary network supply and return water temperature difference: Initialize the population: First, create a set of initial locust positions, each of which represents a possible booster pump variable frequency setting scheme.

[0160] Select reference point: Randomly select the position of a locust ( ): Randomly select the position of a locust from the current population as a reference point. Determine the current optimal position ( ): Find the position of the locust with the smallest objective function value in the current population, that is, the current optimal position.

[0161] Generate new positions: Differential mutation: Generate a new position update direction based on the difference between the current best position and the randomly selected individual position. This step helps explore new areas in the solution space. Apply a scaling factor (F): Use the scaling factor to adjust the step size of the differential mutation to ensure that the new position can fully explore the solution space, but not deviate too far and lose convergence.

[0162] Update position: Compare fitness values: Compare the generated new position with the original position, and select the position with better fitness as the basis for the next iteration. If the new position performs better, replace the original position; otherwise, keep the original position and continue to participate in subsequent iterations. Repeat iteration: Repeat the above steps until the preset stop condition is met (such as reaching the maximum number of iterations or finding a good enough solution).

[0163] In some embodiments of the present application, the updating formula of the scaling factor F is:

[0164] Among them, φ∈[0,T] is a constant.

[0165] Specifically, the scaling factor F is a key parameter that determines how much the new position changes relative to the original position. By adjusting the value of F, you can control the distance that the population members move during the search process: Larger F value: allows larger jumps, helps to quickly jump out of the local optimal solution and increase the global search capability.

[0166] Smaller F value: results in smaller step size, making the search more refined and facilitating deeper exploration in discovered good areas.

[0167] In the improved locust optimization algorithm, the scaling factor F is not a fixed constant, but is dynamically updated according to the number of iterations. This dynamic adjustment mechanism can better adapt to the search needs at different stages: Initial stage: A larger F value is needed to explore the solution space extensively and find potential high-quality solutions.

[0168] Later stage: As the number of iterations increases, the F value is gradually reduced to conduct a more detailed search near the high-quality solutions that have been found, avoid premature convergence and improve the quality of the solution.

[0169] Constant φ: In this context, φ is a constant between 0 and T (maximum number of iterations). It is often used to indicate the relative position of the current iteration in the entire search process. For example, if the current iteration is the 50th iteration and the total number of iterations is 200, then φ=50 / 200=0.25.

[0170] Dynamic adjustment: The scaling factor F is adjusted according to the ratio of the current number of iterations to the maximum number of iterations. This means that at different stages of the search process, The value of F will vary: Initial stage: When the number of iterations is small, φ is small and the value of F is large, which encourages extensive exploration and helps the algorithm find the direction of the global optimal solution.

[0171] From mid-term to late-term: As the number of iterations increases, φ increases and the value of F gradually decreases, making the search more concentrated near the discovered high-quality solutions and improving the quality of the solutions.

[0172] Assume that the improved locust optimization algorithm is being used to optimize the frequency conversion setting of the primary network booster pump of the heat exchange station system: Initialization: Set the maximum number of iterations T=100 and the initial scaling factor F0=0.8.

[0173] Adjustments during the iteration process: At the 10th iteration, φ = 10 / 10 / XMLSchema = 0.1. At this time, according to the update formula, the new F value calculated is larger, which encourages the algorithm to explore the solution space extensively.

[0174] As the number of iterations increases, at the 50th iteration, φ=50 / 100=0.5. At this point, the value of F decreases, but still maintains a certain exploration capability.

[0175] At the 90th iteration, φ = 90 / 100 = 0.9. At this point, the value of F decreases significantly, and the algorithm focuses on performing a refined search near the discovered high-quality solutions to ensure that a high-quality solution is eventually found.

[0176] Results: Through this dynamic adjustment mechanism, the algorithm can widely explore the solution space in the early stage and find potential high-quality solutions; while in the later stage, it can focus on performing refined searches near these solutions, thereby improving the overall search efficiency and solution quality.

[0177] In some embodiments of the present application, in step S3, the locust position update formula of the reverse learning mechanism is:

[0178] in, is the d-dimension of the i-th locust in the t-th iteration, , is the boundary of the locust in the dth dimension, and rand is a random number.

[0179] Specifically, the reverse learning mechanism is a technique to enhance local development capabilities and escape from local optimality. It calculates the "reverse" position of the current individual position, evaluates its fitness value, and selects a better position for update. This strategy increases population diversity and improves the probability of finding the global optimal solution.

[0180] The d-dimension of the i-th locust in the t-th iteration ( ): This refers to the position coordinates of the i-th locust in its d-dimensional position during the t-th iteration. For example, in two-dimensional space, each locust has a two-dimensional position (x and y).

[0181] boundary( and ): Each dimension has an upper bound ( ) and the lower bound ( ), defines the maximum and minimum values ​​allowed for that dimension. These bounds ensure that all positions are within reasonable limits and avoid exceeding the range of practically possible control parameters.

[0182] The concept of reverse position: For each locust individual position, its reverse position can be defined as some form of mapping, usually based on the reflection of the boundary conditions. Specifically, if the current position in a certain dimension is , then its reverse position can be calculated by reflection of the upper and lower bounds.

[0183] Calculation of reverse position: Assume that in a specific dimension d, the current position of the locust is , and the upper and lower bounds of this dimension are and The reverse position is calculated by calculating the symmetric point of the current position relative to the boundary. This method helps the algorithm explore the "other side" of the current position and find possible better solutions.

[0184] Fitness value: The fitness value reflects the goodness of a certain position and is usually determined by the objective function. In this example, the fitness value may be related to the performance indicators of the heat exchange station system, such as adjustment time, overshoot, etc.

[0185] Best update: Once the reverse position is calculated, the fitness values ​​of the current position and the reverse position need to be compared. If the reverse position is found to have better fitness, it is used to replace the original position and brought into the next iteration; otherwise, the current position is continued to be locally developed to find a better solution.

[0186] Assume that the frequency conversion setting of the primary network booster pump of the heat exchange station system is being optimized: Initialize the population: Create an initial set of locust positions, each of which represents a possible booster pump frequency setting scheme.

[0187] Determine boundaries: Set reasonable upper and lower bounds for each dimension. For example, for the booster pump variable frequency setting, the minimum value may be set to 40Hz ( ), the maximum value is 60Hz ( ).

[0188] Calculate reverse position: For each locust, calculate its reverse position in each dimension. For example, if the current position of a locust in a dimension is 50Hz, its reverse position may be 50Hz. Note that this calculation is based on the upper and lower bounds of the specific dimension.

[0189] Evaluate fitness value: Calculate the fitness value of the current position and the reverse position. This involves substituting these two positions into the objective function and evaluating their corresponding system performance indicators.

[0190] Update position: If the fitness value of the reverse position is better than the current position, replace the current position with the reverse position as the basis for the next iteration; otherwise, keep the current position unchanged and continue to participate in subsequent iterations.

[0191] Repeated iteration: Repeat the above steps until the preset stopping condition is met (such as reaching the maximum number of iterations or finding a good enough solution).

[0192] In some embodiments of the present application, in step S3, γ% of individuals are randomly selected from the initial population, with the current control amount increment as the initial value, and the remaining individuals are randomly generated within the constraint range.

[0193] Specifically, randomly select γ% of individuals with the current control amount increment as the initial value γ% selection: The γ% mentioned here refers to randomly selecting a certain proportion (e.g. 10%, 20%, etc.) of individuals from the initial population. This percentage can be determined based on the needs of the specific problem or experimental experience.

[0194] Current control quantity increment: refers to the control quantity change value obtained based on the existing control system or previous optimization results. For example, in a heat exchange station system, this may be the increment used for the most recent adjustment of the booster pump frequency conversion.

[0195] Applied to the initial population: The selected individuals will directly use the current control increment as their initial position. This is based on the assumption that the current control is close to the optimal solution or at least a good starting point, thus providing a good starting point for the search.

[0196] Constraint range: Each variable in each dimension has its upper and lower limits, which define the maximum and minimum values ​​allowed. For example, when adjusting the frequency conversion of the booster pump of the heat exchange station system, there may be a minimum frequency of 40Hz and a maximum frequency of 60Hz.

[0197] Random generation: In addition to the individuals mentioned above that use the current control increment as the initial value, the remaining individuals will be randomly generated within their respective constraints. This means that their positions are completely random and cover different parts of the entire feasible solution space.

[0198] Purpose: In this way, some potential high-quality solutions based on existing knowledge (such as the current control amount increment) are retained, and a large amount of randomness is introduced to ensure that the population has sufficient diversity to better explore the entire solution space.

[0199] Assume that the frequency conversion setting of the primary network booster pump of the heat exchange station system is being optimized, and it is decided to use the improved locust optimization algorithm to perform this task. The initial population size is set to 50 individuals, and γ=20% is selected, that is, 10 individuals will use the current control amount increment as the initial value, and the remaining 40 individuals will be randomly generated within the constraint range.

[0200] Determine the current control increment: Assuming that the booster pump frequency setting in the current system is 50Hz, and the most recent adjustment recommends increasing it to 52Hz, then 52Hz is the current control increment.

[0201] Select 20% of the individuals: Ten individuals are randomly selected from the 50 individuals.

[0202] The positions of these 10 individuals will be set to 52Hz (or the current control increment corresponding to other relevant parameters) as their starting point.

[0203] Randomly generate the remaining individuals: For the remaining 40 individuals, their positions are randomly generated according to the upper and lower limits of each dimension (for example, the lower limit of the booster pump frequency conversion is 40 Hz and the upper limit is 60 Hz).

[0204] These randomly generated positions ensure that the population is evenly distributed throughout the solution space, increasing the likelihood of finding the global optimal solution.

[0205] Start iteration: After all individuals have been initialized, the improved locust optimization algorithm can be used to iterate and gradually optimize the position of each individual until the preset stopping condition is reached (such as the maximum number of iterations or a specific fitness threshold is met).

[0206] In some embodiments of the present application, in step S4, the constraint condition of the control amount increment is:

[0207] Where φ is the constraint vector, and To control the range of increments, define ; and To control the minimum and maximum values ​​of the increment.

[0208] Specifically, when optimizing control systems, especially complex systems like heat exchange stations, the control quantity increment must be subject to certain restrictions to prevent system overload, equipment damage or unstable performance. These constraints define the maximum and minimum ranges within which the control quantity can vary, ensuring the safety and stability of the system.

[0209] Control volume increment ( ): Refers to the adjustment value made to a certain control variable (such as booster pump frequency conversion) in each control cycle. It represents the change from the current state to the next state.

[0210] Constraint vector (φ): This is a vector that indicates whether the control increments in each dimension meet the constraints. In multidimensional problems, each dimension has its corresponding control increment range. For one-dimensional problems, this vector can be simplified to a scalar; for multidimensional problems, it is a vector containing multiple elements.

[0211] Control the range of increments ( and ) This is the minimum value of the control increment, defining the maximum amount the control can be reduced. If the control increment is less than this value, it is considered invalid, as it may cause the system to operate unstable or exceed physical limits.

[0212] This is the maximum value of the control amount increment, defining the maximum amount that the control amount can be increased. Similarly, if the control amount increment exceeds this value, it is also unacceptable because it may cause equipment overload or other operational problems.

[0213] Define the vector ( ) This vector is often used to express consistency conditions in all dimensions. In the case of multiple dimensions, it means that each dimension follows the same constraints. For example, in two dimensions, this vector can be written as , indicating that both dimensions need to adhere to the same minimum and maximum control increment limits.

[0214] Assume that the frequency conversion setting of the primary network booster pump of the heat exchange station system is being optimized, and the following constraints are set: Determine the control quantity increment range: Assume that the minimum increment of the booster pump frequency conversion is =-0.5Hz, which means that in one control cycle, the frequency conversion can only be reduced by 0.5Hz at most. The maximum increment is =0.5Hz, that is, in one control cycle, the frequency conversion can only increase by 0.5Hz at most. Application constraints: In each iteration, the algorithm calculates a new control quantity increment If the calculated Less than or greater than , it needs to be adjusted so that it falls within the allowed range. For example, if the calculated increment is -0.6Hz, it should be adjusted to -0.5Hz; if it is 0.7Hz, it should be adjusted to 0.5Hz. The role of the constraint vector: The constraint vector φ is used to check whether the control amount increment in each dimension meets the above conditions. If the increment in a certain dimension does not meet the conditions, it is adjusted to meet the requirements. For multi-dimensional problems, each dimension needs to be checked and adjusted separately to ensure that the control amount increments in all dimensions are within the specified range. The role of the definition vector ∮: This vector Indicates that all dimensions need to follow the same constraint rules. For example, if there are three dimensions representing different control parameters (such as booster pump frequency conversion, valve opening, etc.), each dimension needs to follow the same and scope.

[0215] A computer-readable storage medium stores a program, which, when executed by a processor, implements the steps in a predictive control method based on improved locust optimization.

[0216] An electronic device comprises a memory, a processor and a program stored in the memory and executable on the processor, wherein when the processor executes the program, steps of a predictive control method based on improved locust optimization are implemented.

[0217] Anything not described in this application can be achieved by adopting or drawing on existing technologies.

[0218] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0219] The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A predictive control method based on improved locust optimization, applied to a heat exchange station system, characterized in that: include: S1: The recursive least square method with forgetting factor is used to identify the parameters of the mass regulation channel of the heat exchange station system, and a controlled autoregressive moving average integral model is established; S2: Based on the controlled autoregressive moving average integral model, generating a predicted output and constructing an objective function through implicit generalized predictive control, wherein the implicit generalized predictive control directly identifies a prediction matrix and a prediction vector; S3: When the control amount increment exceeds the preset constraint range, the objective function is optimized by using an improved locust optimization algorithm, wherein the improved locust optimization algorithm combines a double differential evolution strategy and a reverse learning mechanism, wherein: The double differential evolution strategy updates the locust position through differential mutation, and the mutation basis vector is based on the current optimal position and the random individual position; The reverse learning mechanism updates the locust position by comparing the fitness value of the current position of the locust individual with the reverse position; S4: adjusting the control amount according to the optimization result to achieve the optimal control of the heat exchange station system.

2. The method according to claim 1, characterized in that In step S1, the expression of the controlled autoregressive moving average integral model is: in, It is a differential operator, the input u(k) is the frequency conversion of the primary booster pump, and the output y(k) is the secondary supply and return water temperature difference.

3. The method according to claim 1, characterized in that In step S2, the expression of the objective function is: Among them, n is the prediction time domain, m is the control time domain, λ k is the control weighting constant.

4. The method according to claim 1, characterized in that: In step S3, the locust position update formula of the double differential evolution strategy is: in, x i (t) is the position of the i-th locust in the iteration process, x j (t) is the random position of the locust, The optimal position of , F is the scaling factor.

5. The method according to claim 4, characterized in that The update formula of the scaling factor F is: Among them, φ∈[0,T] is a constant.

6. The method according to claim 1, characterized in that In step S3, the locust position update formula of the reverse learning mechanism is: in, x i d (t) is the d-dimension of the i-th locust in the t-th iteration, ub d , lb d is the boundary of the locust in the dth dimension, and rand is a random number.

7. The method according to claim 1, characterized in that In step S3, γ% of individuals are randomly selected from the initial population, with the current control amount increment as the initial value, and the remaining individuals are randomly generated within the constraint range.

8. The method according to claim 1, characterized in that: In step S4, the constraint condition of the control amount increment is: Where φ is the constraint vector, and To control the range of increments, define ; and To control the minimum and maximum values ​​of the increment.

9. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, the steps in the predictive control method based on improved locust optimization as described in any one of claims 1 to 8 are implemented.

10. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the predictive control method based on improved locust optimization as described in any one of claims 1 to 8 are implemented.

Images