Energy consumption optimization method for wastewater treatment plants based on dissolved oxygen concentration prediction

Abstract

This invention relates to the field of blower speed control technology, and more specifically, to a wastewater treatment plant energy consumption optimization method based on dissolved oxygen concentration prediction. The method includes: collecting multi-dimensional time-series data during the wastewater treatment process, wherein the multi-dimensional time-series data includes DO concentration data for each aeration zone; calculating the global standard deviation of historical DO concentration data and the local standard deviation of DO concentration data within the corresponding local window at the current moment, and using the ratio of the global standard deviation to the local standard deviation as the local fluctuation at the current moment. This invention, through comprehensive exploration weights, can adaptively adjust the amplitude parameters and population diversity of the SMA algorithm, effectively identifying sudden changes in real-world operating conditions and short-term noise interference. This method breaks through the limitations of fixed parameters in conventional algorithms, achieving global precision optimization and energy saving under highly dynamic and complex scenarios such as influent impact, while ensuring effluent quality meets standards and the blower equipment operates stably.

Description

Technical Field

[0001] This invention relates to the field of fan speed control technology. More specifically, this invention relates to a method for optimizing energy consumption in wastewater treatment plants based on dissolved oxygen concentration prediction. Background Technology

[0002] Wastewater treatment is a crucial infrastructure for urban public health and ecological environment protection, characterized by massive energy consumption and high operating costs. In mainstream biological treatment processes such as activated sludge, the aeration system is the core energy-consuming unit, typically accounting for 50% to 70% of the total energy consumption of a wastewater treatment plant. Dissolved oxygen (DO) concentration, as a key control indicator in the aeration process, directly determines the efficiency of microbial degradation and the compliance rate of effluent quality: excessively high DO concentrations lead to energy waste, while excessively low concentrations result in excessive ammonia nitrogen levels in the effluent, posing environmental violation risks. Influenced by the nonlinear coupling of multiple factors such as influent flow rate, organic load, water temperature, and diurnal and seasonal fluctuations, DO concentration exhibits significant non-stationarity and multi-scale dynamic characteristics. Traditional fixed-point aeration control strategies cannot respond to changes in operating conditions in real time, necessitating a dynamic energy consumption optimization method based on accurate DO concentration prediction to achieve refined energy-saving management throughout the entire wastewater treatment process.

[0003] To address the aforementioned aeration energy consumption optimization needs, the Slime Mould Algorithm (SMA) has been introduced into the field of industrial system parameter optimization and energy management in recent years due to its simple structure, lack of pre-training requirements, and excellent global optimization capabilities. SMA uses the foraging behavior of slime molds as a biomimetic prototype, simulating the dynamic process of mycelia moving towards food sources through a sinusoidal adaptive weighting mechanism. It achieves a natural transition between the exploration and development stages, offering advantages such as fast convergence speed, few parameters, and strong interpretability, making it suitable for nonlinear and nonconvex energy consumption optimization scenarios. However, the dissolved oxygen concentration (DO) prediction sequence exhibits significant temporal abrupt changes under unsteady conditions such as influent shock loads and sudden seasonal temperature fluctuations. When directly applying SMA, its weight update strategy relies on a fixed sinusoidal amplitude decay law, failing to perceive the real-time dynamic characteristics of the DO prediction sequence and adaptively adjust the exploration weight allocation of each candidate solution accordingly. This results in insufficient exploration capability under unsteady conditions, leading to local optima and a significant decrease in the accuracy and stability of aeration energy consumption optimization. Summary of the Invention

[0004] This invention provides a wastewater treatment plant energy consumption optimization method based on dissolved oxygen concentration prediction, aiming to solve the problem that when SMA is applied in related technologies, the algorithm has insufficient exploration ability under non-steady-state conditions, gets trapped in local optima, and the accuracy and stability of aeration energy consumption optimization are significantly reduced.

[0005] This invention provides a wastewater treatment plant energy consumption optimization method based on dissolved oxygen concentration prediction, comprising: collecting multidimensional time-series data during the wastewater treatment process, wherein the multidimensional time-series data includes DO concentration data of each aeration zone; calculating the global standard deviation of historical DO concentration data and the local standard deviation of DO concentration data within the corresponding local window at the current moment, and using the ratio of the global standard deviation to the local standard deviation as the local volatility at the current moment; extracting a volatility subsequence composed of all local volatility within a preset time period with the current moment as the endpoint, and calculating the proportion of moments with local volatility greater than 1 in the local volatility subsequence to the total number of moments, as the continuous anomaly proportion at the current moment; constructing a comprehensive exploration weight based on a preset functional relationship between the local volatility and the continuous anomaly proportion; and using the comprehensive exploration weight to adjust the amplitude parameters of the slime mold optimization algorithm, using the adjusted slime mold optimization algorithm to determine the optimal aeration air volume in the aeration tank, and controlling the operation of the blower according to the optimal aeration air volume. Compared to traditional optimization algorithms that rely on fixed parameters, this algorithm can adaptively adjust the amplitude parameters of the slime mold optimization algorithm by sensing the instantaneous local fluctuations and the proportion of continuous anomalies in dissolved oxygen concentration in real time. This effectively enhances the algorithm's global exploration capability under unsteady conditions such as influent shock loads or sudden temperature changes, preventing it from getting trapped in local optima and significantly improving the optimization accuracy and overall system stability of the actual aeration process in wastewater treatment.

[0006] Furthermore, a comprehensive exploration weight is constructed, including: multiplying the difference between the local fluctuation degree and the benchmark value by the proportion of persistent anomalies to obtain an excess dynamic component; and adding the excess dynamic component to the benchmark value to obtain the comprehensive exploration weight. By jointly linearly modeling the instantaneous fluctuation intensity and the duration over time to calculate the comprehensive exploration weight, in application scenarios with relatively mild water quality changes, it not only consumes minimal computational resources but also effectively filters out noise interference such as short-term sensor drift, quickly tracks and smoothly responds to linear fluctuations in daily operating conditions.

[0007] Furthermore, a comprehensive exploration weight is constructed, including: using the local volatility as the base and the proportion of persistent anomalies as the exponent, and performing a power operation to obtain the comprehensive exploration weight. Compared to linear calculation, it plays a crucial damping role in extreme impact scenarios such as sudden heavy rain causing extremely violent fluctuations in water inflow, effectively smoothing the control commands issued to the hardware frequency converter, avoiding the risk of severe surge and burnout of the wind turbine motor, and balancing strong shock resistance and stable equipment operation.

[0008] Furthermore, the method of using the adjusted slime mold optimization algorithm to determine the optimal aeration volume of the aeration tank further includes: after each algorithm iteration, randomly reinitializing a portion of the candidate solutions in the solution space, wherein the proportion of the candidate solutions is proportional to the comprehensive exploration weight. This method, when the algorithm is in a highly dynamic state, forces diversity into the population through proportional global random resampling to prevent premature convergence errors due to changes in the solution space. Under stable conditions, it retains the original progress, avoiding unnecessary interference with the normal development and optimization process.

[0009] Furthermore, determining the optimal aeration volume for the aeration tank involves: within each control cycle, inputting historical multi-dimensional time-series data of a predetermined length prior to the current moment into a trained prediction model, outputting predicted DO concentration values ​​for future control cycles; and using these predicted DO concentration values ​​to optimize the objective function of the adjusted slime mold optimization algorithm to obtain the optimal aeration volume. This method enables the algorithm to proactively optimize the target by incorporating future predicted trends of dissolved oxygen concentration within each control cycle, thereby guiding the aeration system to respond to changes in process requirements in advance and accurately, achieving dynamic and refined energy saving throughout the entire process.

[0010] Furthermore, the optimization process in the objective function of the slime mold optimization algorithm also includes constraint verification. This constraint verification includes ensuring that the predicted DO concentration is within a preset range and that the effluent ammonia nitrogen concentration does not exceed the discharge standards. The mandatory inclusion of environmental constraint verification for dissolved oxygen concentration range and effluent ammonia nitrogen concentration ensures that the final effluent quality of the wastewater treatment plant strictly meets discharge standards under any extreme energy-saving optimization state, effectively mitigating the environmental violation risks for enterprises.

[0011] Furthermore, the method for obtaining the historical DO concentration data is as follows: extract DO concentration data for no less than a preset number of days as historical concentration data.

[0012] Furthermore, the trained prediction model is input, wherein the prediction model is a long short-term memory network model or a neural network model. Leveraging the excellent representational capabilities of long short-term memory networks or neural networks in processing highly nonlinear time-series data, compared to traditional linear prediction models, they can more accurately capture the multi-scale dynamic characteristics of dissolved oxygen caused by the coupling of multiple factors, further solidifying the prediction foundation for energy consumption optimization.

[0013] Furthermore, after acquiring multidimensional time-series data containing dissolved oxygen concentration, the process also includes preprocessing the multidimensional time-series data. The preprocessing includes: filling in missing data using linear interpolation; detecting and removing outliers using the 3σ criterion and replacing the outliers with the mean of adjacent normal values; and performing min-max normalization on all data.

[0014] Furthermore, controlling the blower operation based on the optimal aeration airflow includes: sending the optimal aeration airflow setpoint to the frequency converters of each aeration zone to adjust the blower speed. By directly transmitting the calculated optimal airflow setpoint to the frequency converters of the bottom aeration zone, an automated closed loop is established from software optimization and data prediction to precise operation of physical equipment, truly translating the energy-saving potential of intelligent algorithms into a reduction in the actual energy costs of wastewater treatment plants.

[0015] Beneficial Effects: By extracting the instantaneous local fluctuations and the proportion of continuous anomalies in the time dimension of the dissolved oxygen sequence in real time, a comprehensive exploration weight for joint sensing is constructed. This comprehensive exploration weight can adaptively adjust the amplitude parameters and population diversity of the SMA algorithm, effectively identifying sudden changes in real operating conditions and short-term noise interference. This method breaks through the limitations of fixed parameters in conventional algorithms, and achieves global accurate optimization and energy saving in highly dynamic and complex scenarios such as influent impact, while ensuring that the effluent water quality meets standards and the blower equipment operates stably. Attached Figure Description

[0016] Figure 1 This is a schematic flowchart illustrating an energy consumption optimization method according to an embodiment of the present invention; Figure 2 This is a schematic illustration of an adaptive distribution surface plot of integrated exploration weights according to an embodiment of the present invention. Detailed Implementation

[0017] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0018] To implement the method of this invention, a closed-loop control hardware architecture for optimizing aeration energy consumption in wastewater treatment plants is first required. The specific interaction relationships of this architecture are as follows: Dissolved oxygen (DO) sensors and platinum resistance temperature sensors are deployed in each reaction zone of the aeration tank in the wastewater treatment plant; electromagnetic flow meters and online water quality analyzers are installed in the influent pipeline, where the online water quality analyzer is used to detect the influent COD concentration and the effluent ammonia nitrogen concentration; and smart energy meters are installed in the aeration system's power distribution cabinet. All the aforementioned sensors and energy meters are connected to a data acquisition controller via an RS485 bus or industrial Ethernet, which is further connected to an industrial control computer. The industrial control computer sends control commands to the frequency converters of the blowers in each aeration zone via the communication bus. The frequency converters adjust the motor frequency to change the blower speed, thereby regulating the aeration airflow.

[0019] This embodiment provides two examples of the complete workflow for implementing closed-loop control for energy consumption optimization in the daily operation of a wastewater treatment plant. Example 1 is shown below.

[0020] like Figure 1 As shown, S101: Data acquisition and preprocessing.

[0021] In this embodiment, the data acquisition controller presets a sampling interval and synchronously collects data on DO concentration, influent flow rate, influent COD, effluent ammonia nitrogen, water temperature, and aeration energy consumption, which are then integrated into the industrial control computer to form a multidimensional time-series dataset. The preset sampling interval is 1 minute.

[0022] For the collected data, if data loss due to communication interruption is detected, linear interpolation is performed to fill in the missing data using valid data from before and after the missing value. Then, the 3σ criterion is used to detect and remove outliers introduced by sensor drift, replacing the outliers with the mean of adjacent time points. Finally, all dimensions of the collected data are subjected to min-max normalization to eliminate the influence of dimensional differences on subsequent calculations.

[0023] S102: Calculate the local volatility at the current moment and determine the exploration amplitude parameters for this iteration.

[0024] In time-series driven optimization problems, the exploration capability required by an algorithm is directly related to the complexity of the solution space, which in turn is determined by the dynamic intensity of the current sequence. For aeration energy consumption optimization driven by DO concentration prediction, the more severe the local fluctuations in the DO sequence, the more complex the nonlinear mapping relationship between aeration rate and DO concentration, the more dispersed the distribution area of ​​the optimal solution, and the greater the exploration amplitude required by the algorithm. Based on this premise, the key to determining whether SMA (Sequencing Analysis Method) needs to strengthen its exploration lies in accurately measuring the intensity of local fluctuations in the DO sequence at the current moment.

[0025] From the perspective of time series analysis theory, if the DO series satisfies the wide stationarity condition on a long-term historical scale, then the fluctuation amplitude within any local time window should statistically be consistent with the global historical level, i.e., the ratio of the local standard deviation to the global standard deviation should approach 1. When disturbances such as inflow shock loads or sudden temperature changes break this stationarity, the fluctuation amplitude within the local window will exceed the global historical benchmark, and the ratio will be greater than 1, with the degree of deviation being positively correlated with the intensity of the disturbance. Therefore, using this ratio to measure the non-stationarity of the current operating condition has a natural theoretical basis: the stationarity assumption holds when the ratio equals 1; the larger the ratio is than 1, the more significant the deviation from stationarity, the more drastic the dynamics of the current operating condition, and the more complex the solution space of the sequence-driven model, requiring the algorithm to conduct more extensive exploration. This boundary is naturally given based on the wide stationarity assumption itself, without relying on any artificially set empirical thresholds.

[0026] Based on the above analysis, at the beginning of each iteration, the industrial control computer extracts historical DO concentration data from the database for at least 30 days and calculates the global standard deviation as a long-term stable benchmark. At the beginning of each control algorithm iteration, the industrial control computer extracts a DO concentration subsequence with a length of 30 sampling points ending at the current time and calculates its local standard deviation. The ratio of the local standard deviation to the global standard deviation is used as the local volatility at the current time. When the local volatility is greater than 1, the current local volatility exceeds the historical normal range, and the local volatility directly replaces the exploration amplitude parameter of this iteration, so that the exploration range expands proportionally with the dynamic degree of the working condition; when the local volatility is not greater than 1, the current local volatility does not exceed the historical benchmark, the exploration amplitude remains unchanged, and the algorithm maintains a fine development state. The larger the local volatility, the more fully the exploration amplitude expands, and the stronger the algorithm's ability to cross the local optimal barrier under highly dynamic working conditions; when the working condition is stable, the local volatility naturally falls back to around 1, the amplitude does not expand, and the development accuracy is maintained.

[0027] S103: Calculate the proportion of persistent anomalies at the current moment.

[0028] The amplitude dynamic adjustment mechanism in step S102 relies on instantaneous local fluctuations, which can respond to changes in the operating conditions of the DO sequence on a time-by-time basis. However, this instantaneous response method has an inherent limitation: under the influence of factors such as bubble disturbances in the inlet pipe and brief sensor drift, the DO sequence will experience noisy abrupt changes, causing the local fluctuations to rise sharply in a very short time and then fall back quickly. If a large amplitude expansion is triggered at this time, the algorithm will make an ineffective exploration response to the noise, consuming iterative resources without generating optimization benefits. According to statistical process control theory, meaningful changes in operating conditions are usually manifested as process variables continuously deviating from the normal range over a period of time, while noise interference manifests as a brief anomaly followed by rapid recovery. Therefore, instantaneous local fluctuations alone cannot effectively distinguish between real operating condition disturbances and noise interference; a time persistence dimension needs to be introduced on top of instantaneous perception for differentiation.

[0029] Specifically, the industrial control computer extracts a historical local volatility subsequence with the current time as the endpoint and a length of a preset number of sampling points, where the preset number of sampling points is 120. The number of times the local volatility is greater than 1 among these preset number of sampling points is counted, and this number is divided by the total number of times to obtain the proportion of persistent anomalies at the current time. When the proportion of persistent anomalies approaches 0, it indicates that the recent period has been mostly stable, and the current increase in local volatility is more likely due to transient noise; when the proportion of persistent anomalies approaches 1, it indicates that the recent period has been continuously abnormal, and the current local volatility reflects actual operating condition disturbances.

[0030] Based on the proportion of persistent anomalies, active management of SMA population diversity is implemented: after each iteration, the proportion of persistent anomalies is directly assigned to the global random reset ratio parameter of the candidate solutions in the slime mold optimization algorithm. When the proportion of persistent anomalies is high, it indicates that the algorithm is in a high-dynamic state for a long time, and there is a risk of losing diversity in the original candidate solutions during the continuous expansion of the search process. At this time, a large proportion of candidate solutions are globally randomly resampled in the solution space to force the injection of diversity into the population and prevent the algorithm from prematurely converging to an incorrect local optimum in the dynamically changing solution space. When the proportion of persistent anomalies is low, the operating conditions are mainly stable or with short-term disturbances. Only a very small number of candidate solutions are reset to preserve the existing convergence progress of the population and avoid interfering with the normal development process.

[0031] S104: Construct comprehensive exploration weights.

[0032] To address the coordination deficiencies in steps S102 and S103 when processing the dynamic features of the DO sequence, namely the problem that the algorithm over-responds to transient noise and disjointed actions when faced with continuous perturbations due to the independence of the instantaneous intensity and duration dimensions, it is necessary to jointly model the instantaneous intensity and duration.

[0033] The modeling process is as follows: Excess dynamic components where local volatility exceeds the baseline value are considered valid signals to be responded to. The proportion of persistent anomalies is used as the time-dimensional confidence weight of this signal. By multiplying the two and adding them back to the baseline value, a comprehensive exploration weight is constructed without introducing additional parameters, where the baseline value is 1. This construction method achieves mutual constraints on exploration behavior: when the proportion of persistent anomalies is low, even if instantaneous fluctuations are severe, the excess components will be effectively suppressed, causing the comprehensive exploration weight to approach 1 to avoid noise interference; only when both the intensity and duration of fluctuations reach high levels will the comprehensive exploration weight expand significantly, thereby triggering a strong global exploration. In intermediate conditions, the weight ensures the smoothness of algorithm state transitions through a linear transition.

[0034] Then, the calculated comprehensive exploration weight is used to replace the fixed sinusoidal amplitude parameter in the original SMA position update formula, and the candidate solution reset ratio in step S103 is uniformly scaled using the comprehensive exploration weight. This coordinated adjustment mechanism ensures that the expansion of exploration amplitude and the injection of population diversity operate under the same weight system, eliminating the inconsistency when the two dimensions operate independently. The remaining position update rules and slime mold probability allocation mechanism of SMA remain unchanged, with modifications limited to the source of the exploration amplitude parameter, ensuring minimal invasiveness of the improved scheme and complete preservation of the algorithm structure.

[0035] S105: Based on a comprehensive exploration weight adjustment algorithm, it performs energy consumption optimization closed-loop control.

[0036] Within each control cycle, the industrial control computer inputs the historical DO concentration sequence, influent flow rate, COD, and water temperature data from the previous 60 minutes into a pre-set DO concentration prediction model, such as an LSTM or neural network model, and outputs the predicted DO concentration for future control cycles. The improved SMA algorithm is then activated: the blower airflow setpoint for each aeration zone is used as the candidate solution, i.e., the optimization variable. The position of each candidate solution is updated according to the comprehensive exploration weights in step S104. Candidate solutions exceeding the upper and lower limits of the airflow hardware are subject to boundary reflection processing. At the end of each iteration, based on the proportion of persistent anomalies determined in step S103, candidate solutions of the corresponding proportion are globally and randomly regenerated within the solution space. The candidate solutions are substituted into the objective function, which aims to minimize total aeration energy consumption, and constraint verification is performed. The constraint verification requires that the predicted DO concentration value be between 2 and 4 mg / L, and the effluent ammonia nitrogen concentration not exceed the discharge standard. The historical best solution is then updated. After reaching the maximum number of iterations, the industrial control computer outputs the optimal air volume setting value and sends the optimal air volume setting value to the frequency converter controllers of each aeration zone. The frequency converter controllers are used to adjust the speed of the blower, thereby completing one closed-loop control.

[0037] In Example 2, the hardware architecture and steps S101, S102, S103, and S105 are completely identical to those in Example 1. The difference lies in the mathematical construction of the comprehensive exploration weights in step S104. The reason for this is that the method in Example 1 is suitable for situations with relatively mild water quality changes. In this case, linear calculation is simple, computational resources are minimal, and linear changes can be quickly tracked, which is sufficient for daily needs. However, when faced with non-linear and drastically fluctuating influent, linear weights can cause the calculated airflow command to fluctuate wildly in a short period. The frequency converter will frequently control the blower motor to accelerate and decelerate rapidly, which can easily burn out the motor or damage the blower impeller. Therefore, Example 2 is provided for some wastewater treatment plants with extremely severe dynamic non-linearity, employing exponential calculations to enhance the algorithm's response sensitivity.

[0038] Specifically, the industrial control computer uses a power function to construct a nonlinear weight mapping, that is, using the local volatility as the base and the proportion of persistent anomalies as the exponent to construct a comprehensive exploration weight. When the proportion of persistent anomalies approaches 0, which conforms to a noisy operating condition, any base raised to the power of 0 approaches 1, and the comprehensive exploration weight is locked at the baseline value; when the proportion of persistent anomalies approaches 1, which represents a real shock, the comprehensive exploration weight exponentially approaches the local volatility itself. This variant can provide a smoother algorithm transition under extreme operating conditions.

[0039] Finally, to enable those skilled in the art to more clearly understand how this method avoids ineffective exploration and crosses local optima, the system action logic in two actual use scenarios of the embodiments is given below.

[0040] In Example 1, under the scenario of stable normal operation, the influent flow rate is stable and the DO concentration fluctuates little. The local fluctuation degree calculated in step S102 is approximately 1, and the continuous anomaly ratio calculated in step S103 is approximately 0. Substituting these values ​​into the calculation in step S104, the comprehensive exploration weight is equal to 1, and the reset ratio is 0. The improved SMA algorithm maintains the original fine-tuned development state, fine-tuning the airflow in small steps near the current optimal solution to ensure stable effluent and save energy.

[0041] In Example 2, the scenario involves a sudden and intense inflow load caused by heavy rainfall, such as a sudden downpour with extremely drastic fluctuations in water level. The data collected by the industrial control computer shows a local volatility spike to 6.0 and a sustained anomaly rate of 0.8. When the system operates in the linear weighted mode of Example 1, the calculated comprehensive exploration weight is linearly amplified to 5.0, resulting in a very aggressive response. The algorithm instantly abandons the original local optimum, and the blower's frequency converter receives the command, causing the blower speed to surge rapidly within a short period. This method can extremely quickly increase the aeration rate, effectively suppressing the impending exceedance of ammonia nitrogen. However, the trade-off is excessive blower acceleration, high short-term energy consumption, and a certain risk of overshoot. When the system operates in the exponential calculation mode of Example 2, the calculated comprehensive exploration weight is 5.19. Faced with the same local fluctuation of 6.0 extreme fluctuation, the exponential formula played a damping role, compressing the nonlinearity of the weight to 4.19, making the command curve sent from the industrial control computer to the frequency converter smoother, avoiding the severe surge of the blower motor, and achieving the dual purpose of shock resistance and smooth control.

[0042] like Figure 2 As shown in the figure, the function surface of the integrated exploration weight continuously changing with local volatility and the proportion of continuous anomalies in this invention is illustrated. This weight model has self-consistent boundary behavior: the weight is locked at 1 under stable conditions; the weight is amplified proportionally with local volatility under strong and continuous shocks; and the weight returns to near 1 under transient noise due to the suppression of the proportion of continuous anomalies.

Claims

1. A method for optimizing energy consumption in wastewater treatment plants based on dissolved oxygen concentration prediction, characterized in that, include: Collect multidimensional time-series data during the wastewater treatment process, wherein the multidimensional time-series data includes DO concentration data of each aeration zone; Calculate the global standard deviation of historical DO concentration data and the local standard deviation of DO concentration data within the corresponding local window at the current moment, and use the ratio of the global standard deviation to the local standard deviation as the local volatility at the current moment; Extract a volatility subsequence consisting of all local volatility values ​​within a preset time period ending at the current time. Calculate the proportion of times when the local volatility value is greater than 1 in the local volatility subsequence to the total number of times, and use this as the proportion of continuous anomalies at the current time. A comprehensive exploration weight is constructed based on the preset functional relationship between the local volatility and the proportion of persistent anomalies; The integrated exploration weight is used to adjust the amplitude parameters of the slime mold optimization algorithm, the adjusted slime mold optimization algorithm is used to determine the optimal aeration volume of the aeration tank, and the blower is controlled to operate according to the optimal aeration volume.

2. The wastewater treatment plant energy consumption optimization method based on dissolved oxygen concentration prediction according to claim 1, characterized in that, Construct a comprehensive exploration weight, including: The difference between the local volatility and the benchmark value is multiplied by the proportion of persistent anomalies to obtain the excess dynamic component. The excess dynamic component is then added to the benchmark value to obtain the comprehensive exploration weight.

3. The wastewater treatment plant energy consumption optimization method based on dissolved oxygen concentration prediction according to claim 1, characterized in that, Construct a comprehensive exploration weight, including: The comprehensive exploration weight is obtained by exponentiation of the local volatility as the base and the proportion of persistent anomalies as the exponent.

4. The wastewater treatment plant energy consumption optimization method based on dissolved oxygen concentration prediction according to claim 1, characterized in that, The step of using the adjusted slime mold optimization algorithm to determine the optimal aeration air volume for the aeration tank also includes: After each algorithm iteration, a subset of candidate solutions is randomly reinitialized in the solution space, and the proportion of the subset of candidate solutions is proportional to the comprehensive exploration weight.

5. The wastewater treatment plant energy consumption optimization method based on dissolved oxygen concentration prediction according to claim 1, characterized in that, Determining the optimal aeration air volume for the aeration tank includes: Within each control cycle, historical multidimensional time-series data of a set length prior to the current moment is input into the trained prediction model, which outputs the predicted DO concentration for future control cycles. The predicted DO concentration is then used to optimize the objective function of the adjusted slime mold optimization algorithm to obtain the optimal aeration airflow.

6. The wastewater treatment plant energy consumption optimization method based on dissolved oxygen concentration prediction according to claim 5, characterized in that, In the process of finding the optimal objective function of the slime mold optimization algorithm, the constraint condition verification is also included. The constraint condition verification includes that the predicted value of DO concentration must be within a preset range and the ammonia nitrogen concentration of the effluent must not exceed the discharge standard.

7. The wastewater treatment plant energy consumption optimization method based on dissolved oxygen concentration prediction according to claim 1, characterized in that, The method for obtaining the historical DO concentration data is as follows: extract DO concentration data for no less than a preset number of days as historical concentration data.

8. The wastewater treatment plant energy consumption optimization method based on dissolved oxygen concentration prediction according to claim 5, characterized in that, Input the trained prediction model, wherein the prediction model is a long short-term memory network model or a neural network model.

9. The wastewater treatment plant energy consumption optimization method based on dissolved oxygen concentration prediction according to claim 1, characterized in that, After obtaining multidimensional time-series data containing dissolved oxygen concentration, the method further includes preprocessing the multidimensional time-series data. The preprocessing includes: filling missing data using linear interpolation; detecting and removing outliers using the 3σ criterion and replacing the outliers with the mean of adjacent normal values; and performing min-max normalization on all data.

10. The wastewater treatment plant energy consumption optimization method based on dissolved oxygen concentration prediction according to claim 1, characterized in that, Controlling the blower operation based on the optimal aeration air volume includes: The optimal aeration air volume setting is sent to the frequency converter controller of each aeration zone to adjust the speed of the blower.

Images